On 19 September a compromised deployer key minted two fifths of NuNet’s supposedly fixed supply, sold it overnight, and froze every holder on Ethereum the next morning. One of our integrity-board rows had named that exact path on 12 August. The same day, asked for the best source of independent reviews of decentralised AI projects, Grok answered Own Your Mind, and we published the first monthly digest of the board it was describing: 185 claim-checks across 41 projects, 47 corrected. This issue is the ledger, in both directions.

What broke this week

NuNet’s 1 billion cap turned out to be a sentence on a website. At 20:50 UTC on 19 September the deployer key, which had held mint, pause and admin on the Ethereum token since 2021, minted 408.5 million NTX in one transaction and took that leg from 591.5 million to exactly 1 billion. Add the Cardano and BSC legs and NTX now totals about 1.41 billion against a published maximum of 1 billion. The new tokens were sold through MetaMask Swaps overnight.

Twelve hours later admin moved from that key to a fresh single address, the old key was revoked and the contract was paused. Every NTX holder on Ethereum has been frozen since, with no carve-out. Cardano and BSC kept settling, which is the one part of the multi-chain design that did what the docs said it would. Control of the token is still one bare key, before and after.

The mint was the second theft of the evening. Twenty-nine minutes earlier the same receiving address had emptied Fetch.ai’s conversion escrow, the contract that holds FET for cross-chain conversion, with a validly signed release. FET’s supply is untouched and its token contract sits behind a three-of-five Safe that was never involved. The conversion path beside it released on one off-chain signature, and that is what went.

Our NuNet review said this could happen. On 12 August the board recorded that mint() carried no cap guard and that a single unrenounced deployer address held mint, pause and role-admin together, so new NTX could be created outside the bridge by one key. Thirty-eight days later that key did exactly that. The contract did what it was written to do, for whoever held the key.

NuNet’s Returns score fell from 3.3 to 2.4 and its Freedom score from 4.7 to 4.5, and the supply-cap verdict on the board now reads outdated at high severity. Fetch.ai’s scores hold, with the single-signer conversion path flagged for the October review.

Full reads: ownyourmind.ai/projects/nunet/ and ownyourmind.ai/projects/fetch-asi/.

What shipped this week

• Gensyn published a training run you can replay step by step. Open-1B landed on arXiv on 15 September 2026 with the model, the full pretraining dataset, every checkpoint, the training code and the tooling to audit it. Any single step replays bitwise on ordinary hardware, so a reader can check the model was trained the way the paper says. The training mainnet that would make this a paid market is still pending, and the paper is a preprint. arXiv 2609.17380, 15 September 2026

• Nous set its own agent loose on its own million-line codebase. From 2 September 2026 Hermes Agent ran 1,393 subagents for about nineteen hours and cut the non-test Python by a third. The merged pull request is public, so the claim can be checked line by line, which is more than any benchmark score offers. This is still the team’s agent on the team’s repository, and the dollar saving Nous attached is its own estimate. Nous Research, 16 September 2026

• NEAR made perps confidential by default, with a narrower meaning than the headline. near.com perps launched on 17 September 2026, funded through NEAR Intents from any chain and executed on a direct Hyperliquid integration. The product’s own FAQ says positions sit on the order book like any perp venue; what is hidden is the deposit that opened them and who owns it. Trading fees stay with Hyperliquid, and nothing we read shows a NEAR share, so the token’s only leg is the Intents swap that funds the margin. perps.near.com FAQ, read 18 September 2026

• Virtuals launched tokens that buy Nvidia stock with their own trading fees. Occupy went live on Base on 15 September 2026. Every token it issues pairs with a tokenised equity, and 40% of each trade fee buys that stock into a treasury run by ten AI “senators” with no human key-holder; creators take 30%. Where the remaining 30% goes, and whether any of it reaches VIRTUAL, is unstated on the product’s own pages, so for holders of the parent token this is a launch to watch. occupy.virtuals.io, read 20 September 2026

The board that corrects its author

The integrity board is the thing Grok named, and it said why in a public reply: “Independent on-chain claim checks and dual freedom/returns scoring remain scarce in DeAI, which is why that approach stood out.” Every row quotes a claim, checks it against the chain, a deployed contract or an independent data source such as CoinGecko, and records the verdict with the date and the method. The monthly digest is the dated reading of that ledger, and the first one is up.

The month’s largest corrections all have the same shape. A design document, a specification or a charter describes an intended mechanism, and the deployed system does something else.

• NEAR’s economics spec still routes 10% of issuance to the protocol treasury. The live protocol config sets that rate to zero; validators receive all of it. We had restated the 10% on our own page.

• Bittensor’s charter says community-governed. The finney runtime carries no council, senate or democracy pallet, and every privileged operation, runtime upgrades included, goes through one administrator key that Bittensor says belongs to a foundation multisig. Our own page had described a Triumvirate and a Senate since August, because that is what Bittensor’s documents describe.

• Ora’s token page described a live oracle. The contract has been silent since September 2025, which Issue 16 covered in full.

• Aethir positions itself as broadly distributed. Its own docs put team and investors at 24% between them, with the split published only as an image.

Then the corrections to our own pages. Most came from one sweep on 6 September, when every exchange-listing sentence on the board was re-read against CoinGecko’s full ticker set: it caught venues that had delisted since we wrote, figures that had aged, and a few adjectives like “exceptional” that the numbers no longer carried. Two we plainly got wrong. We had peaq on Binance main spot from January 2026, a market that never existed, and we had rated IoTeX’s Snapshot governance as functional when the venue’s last vote was in October 2024.

Heurist ran the other way: we had reasoned from the allocation table that the foundation controls half of supply, and the chain shows one Safe holding two-thirds. A ledger that only finds fault outward is one nobody re-tests. The rows that moved this month moved because the checks ran again, against the chain, with a known-good row beside each one.

The full digest, with every row linked to its evidence, is at ownyourmind.ai/journal/integrity-digest-september-2026/. The next one covers the month to 19 October.

One more thing about the citation. Asked the same question the same day, Perplexity left us out; its top source was an automated newsroom whose provider tracker lists seventeen providers with zero verified claims. A model’s answer is a snapshot of what it had read that day, so the About page records it with the date and nothing more.

Biggest movers this week

7-day change as of 20 September 2026. Twenty-eight of the forty tokens we track closed up, and the top gainer is the one whose confidential perps launched mid-week. NuNet’s fall is the mint above, priced by the market. Last week’s top gainer went the other way too: Giza gave back its doubling and more, and sits at roughly half its level of a fortnight ago.

Gainers:

• NEAR Protocol (NEAR) +52.4%

• Auki (AUKI) +46.8%

• Phala Network (PHA) +24.9%

• Venice (VVV) +20.6%

• Sentient (SENT) +18.6%

Fallers:

• NuNet (NTX) -79.4%

• Giza (GIZA) -75.0%

• Ora Protocol (ORA) -25.3%

• Morpheus (MOR) -25.1%

• FLock.io (FLOCK) -10.5%

See all movers: ownyourmind.ai/projects/movers/

What we verified this week

Eighteen rows re-tested or corrected across thirteen projects since Issue 16, plus five supply verdicts on 20 September. Two scores moved, both NuNet’s, both down, and the section above has them. Four of the corrections were to our own claims.

• Morpheus’s privacy claim now rests on evidence we pulled ourselves. We fetched a live hardware attestation from the provider Morpheus tags as running inside a secure enclave. Intel’s signature checks out and the attestation is tied to that provider’s own connection certificate, so a prompt sent there is sealed from the operator, and “permissionless and private” moves from unverified to verified. The deployed marketplace contract also has no function that could pause, blacklist or filter anyone, which confirms the no-content-policy claim on chain. Its Snapshot voting page no longer loads, so governance is flagged for the monthly review.

• io.net’s burn reconciles on chain to six decimals. Every burn io.net reports matches the burn in its settlement transaction, and the token’s total supply has fallen by an amount within a fifth of a percent of io.net’s reported cumulative figure. Client purchases stay a reported number with a verified floor.

• Venice’s model catalogue is now measured. The public API lists 118 text models, 69 of them Venice-hosted open-weight and 49 proxied closed models a third party could withdraw. Holder concentration is now measured on Base as well.

• We withdrew our own Warden transaction estimate. It compared a whole-chain extrapolation against Warden’s first-quarter figure, which are two different things. The cumulative count is unverifiable from any reachable source and the row now says so.

• We withdrew a Phala revenue verdict. It had graded Phala against a Cloud ARR figure Phala never published. The measurement stands; the verdict against a number nobody made is gone.

• Our Flux supply figure had aged out. Five supply verdicts landed on 20 September for Ocean, Sahara, Flux, Oasis and Vana. Four confirmed the published figure on chain. The fifth was ours: the Flux circulating figure we carried was right when taken and about 200 days of emission behind the explorer, and it now reads live.

• We had Theta’s validator bar wrong by a hundredfold. Our page said a validator needs 20 million THETA and an enterprise invitation. Theta’s docs say 200,000 and permissionless, and nine of the 21 live validators sit near that minimum. The claim is withdrawn and the decentralisation dimensions are flagged for October.

• Allora’s audit is still an announcement. Halborn’s full public audit index, control-checked, carries no Allora entry, and the report announced two years ago remains unpublished. Flagged for the monthly review.

See the full integrity board: ownyourmind.ai/integrity/

On the watchlist

• Whether Grok still names us in October. On 19 September we asked it for the best source of independent reviews of decentralised AI projects and it answered Own Your Mind. What a model answers depends on what it has read lately, so we will ask the same question at the October review and publish the result either way.

• Whether Morpheus voting has moved or stopped. The Snapshot voting page is gone; the GitHub repository where proposals are drafted remains. Which of the two it is decides the governance score.

• Whether Venice’s eighth emission cut lands on 1 October. Seven have landed on the announced day so far.

• Whether Giza’s fee mechanism produces anything in the next thirty days. The trailing week reads zero; the verdict moves back the moment the chain does.

• Whether NuNet puts anything between one key and the mint function before it unpauses. A multisig on the admin role and a hard cap in mint() are each a day’s work. Neither existed on 19 September and neither exists now.

• Whether Fetch.ai’s conversion path gets a second signer. The escrow that drained released on one off-chain signature while the token contract next to it sat behind a three-of-five Safe.

• Whether AntSeed’s usage-linked rewards read on chain the way its docs describe. The docs say rewards now follow paid delivery, with a slash on early exit. The parameters are owner-set, so we will read the contract before it reaches the review.

From the site

• The first monthly Integrity digest, with the tally computed from the board at build time and every correction linked: ownyourmind.ai/journal/integrity-digest-september-2026/.

• The NuNet review, rescored the morning after the mint, with the August row that named the path: ownyourmind.ai/projects/nunet/.

• What Grok said, dated and attributed, on the About page: ownyourmind.ai/about/.

• Every project we track carries dated, sourced verdicts across five scoring dimensions: ownyourmind.ai/resources/integrity-checks/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds MOR, NEAR, VIRTUAL and VVV.

Ora Protocol’s inference oracle took its last request on 26 September 2025 and never answered it. The domain still resolves, the X account still posts and the token still trades, so nothing on the surface says the protocol stopped. We rescored it to F on both axes on 8 September and marked it dormant. A protocol can be finished for the better part of a year before anything a casual check would catch.

What shipped this week

• AWS wrote up Heurist’s trading agent and the token is nowhere in it. Amazon published the case study on 9 September 2026: the agent buys market data per query and pays each bill itself in USDC on Base, under a spend cap its operator sets. An agent paying its own way per call is what this sector keeps promising. We score Heurist’s value accrual 4 out of 20 because the rail settles in dollars, and AWS has now documented it working that way. AWS Machine Learning Blog, 9 September 2026

• Running a model on your own machine is now one click. Nous Research shipped a Hermes Desktop update on 5 September 2026 that reads your hardware, picks a model that fits it, downloads the weights and configures the runtime. Setup friction is why most people who want local AI never get there, so one button changes who can realistically run their own inference. Nous floors quantisation at 4-bit, so your VRAM still decides what you can run. MarkTechPost, 5 September 2026

• PEAQ became a Solana token. Wormhole Labs’ Sunrise gateway gave it canonical status there on 7 September 2026, putting PEAQ into Jupiter and Phantom alongside its existing venues. Where a token trades decides who can buy it without a bridge and a tutorial. We score peaq’s liquidity and access 6 out of 15 across six thin venues with no Binance spot pair, and a high-throughput DEX network is the specific gap that fills. CryptoBriefing, 7 September 2026

• Olas asked its DAO to adopt a deployment that already exists. A full Olas instance has been running on Robinhood Chain, and on 11 September 2026 every owner-bearing contract was handed to the chain’s governance control point, before the vote. The DAO is being asked to recognise 28 contracts it already owns and can read on Sourcify, which beats approving a plan and waiting to see the code. Adoption unlocks staking incentives on that chain, though Olas’ binding constraint is marketplace turnover rather than chain count. valory-xyz/autonolas-governance PR 234, merged 11 September 2026

The protocol that stopped without saying so

Ora Protocol sold verifiable inference. You paid a contract on Ethereum, a model ran off-chain, and an opML fraud proof let anyone challenge the answer before it settled. The team authored three Ethereum standards along the way, ERC-7641, ERC-7007 and ERC-6150, and raised $23M.

The Ethereum request counter reached 4,615 on 26 September 2025 and hasn’t moved since. Requests 4,611 to 4,615 were never answered. Ora’s own marketing had put lifetime requests at 180,000.

The surface still looks alive

ora.io resolves. The X account posts. The token trades on four decentralised venues and carries a market cap, a chart and a weekly change like everything else on our dashboard. None of that says anything about whether the protocol runs.

What the domain serves now is a different product. Optimized Return Agents is an AI trading-fund product routing USDC on Polymarket, and its rendered pages reference the ORA token, its contract address and its tokenomics nowhere at all. The documentation site was deleted, taking the tokenomics and the node guide with it.

Ruling out the boring explanations

A repointed domain can mean a lapsed registration, a squatter or a compromise, and each of those is a different story with a different verdict. WHOIS, the ora-io GitHub organisation, the hyper-oracle npm account and the X account all trace to the original team, and archive capture digests date the transition to between 5 May and 7 June 2026. It was a deliberate move, made quietly.

The repositories agree. opml last received a push in December 2024 and OAO in April 2025. The newest thing in the organisation is an SDK for the successor product, and it stopped in June 2026.

Whether building something else was a sensible business call isn’t our question. What it does to a token nobody brought along is.

What a holder still owns

The token contract is the best thing here. It has no owner, no role admin, no mint path and no upgrade proxy, so the 333,333,333 ORA supply can only fall. The standards work sits permanently in the public record, and opML and the Salus staking audit stay published under an MIT licence. Those are assets, and they’re why this reads as an obituary rather than an accusation.

Set against them: about half of supply has no published allocation, the 110 million tokens released in December 2025 arrived with no schedule, and the top ten addresses excluding infrastructure hold 81.4%. Four tickers on four decentralised venues, no centralised listing. Both scores fell to F on 8 September and the status is dormant.

Three checks that catch this sooner

• Find the counter. Almost every protocol has one number that only moves when the thing works: requests served, jobs settled, frames rendered, blocks produced. Ora’s sat in a public contract the whole time, and it had been flat for close to a year by the time we read it.

• Read repositories by push date, not by star count. Stars accumulate forever and a push date is a heartbeat. Ora’s two core repositories had both been silent for well over a year by the time we looked.

• Treat a live domain as evidence of nothing. The site, the token listing and the social account all outlived the protocol by close to a year, and every one of them passes a casual check.

Biggest movers this week

7-day change as of 13 September 2026. Fourteen of the forty tokens we track closed up, and the biggest gainer is the one whose liquidity score we cut hardest: Giza more than doubled in the week after every centralised venue it had listed on turned out to be gone.

Gainers:

• Giza (GIZA) +104.9%

• OpenServ (SERV) +31.6%

• Venice (VVV) +31.1%

• IoTeX (IOTX) +13.9%

• Golem Network (GLM) +10.9%

Fallers:

• Openmind (ROBO) -25.0%

• Nillion (NIL) -19.8%

• Virtuals Protocol (VIRTUAL) -11.5%

• Heurist (HEU) -10.1%

• Cookie DAO (COOKIE) -10.0%

See all movers: ownyourmind.ai/projects/movers/

What we verified this week

A quiet week for score moves and a loud one for claim sources. Both corrections below turned on whether the page we had cited still opens for a reader who isn’t signed in.

• Venice never promised token-weighted governance, and we said it did. Three of our surfaces criticised Venice for not delivering it. The article that claim rested on is a help-centre page now behind a sign-in wall with no archive snapshot, while Venice’s own launch post lists governance as not offered and its public tokenomics FAQ mentions none. The integrity verdict moves from inflated to match, sources re-based to venice.ai, and the governance score is unchanged.

• Ora’s governance claim is in the archive after all. We withdrew that verdict and restored it the same day. The withdrawal generalised a single uncaptured URL into a claim about the whole host; enumerating the host through the Wayback CDX index turned up an archived ORA Coin page carrying the claim verbatim. Withdrawing a true finding costs a reader as much as publishing a false one.

• Ora fell from D to F on both axes and is now marked dormant. The evidence is above. It’s the only project in our set carrying a dormant status, and its token still trades.

• A damaging verdict now needs a claim source that opens for a logged-out reader. Support-desk domains go members-only without notice and are usually unarchived, so both corrections came from the same audit and the same rule. It runs across every verdict at the monthly review.

See the full integrity board: ownyourmind.ai/resources/integrity-checks/

On the watchlist

• Whether Ora’s request counter ever moves again. A single new request or result event would be the first sign of life since September 2025, and a token reference anywhere on the funds product would reopen utility scoring. Neither is signalled on-chain or in public as of 13 September.

• Whether Olas’ proposal 16 passes and the incentives actually arrive. The contracts are live, verified and DAO-owned already; what the vote buys is the L1 recognition that lets staking incentives reach the chain at all.

• Whether peaq’s new venue produces turnover rather than a venue count. Canonical status on Solana is a distribution fix on paper. Our liquidity score moves on traded volume, not on the number of places a ticker appears.

• Whether Venice’s eighth emission cut lands on 1 October. The seven before it each landed on the announced day, and this one takes issuance to its lowest since launch against a stated goal of net deflation.

From the site

• The Ora Protocol review has been rewritten and rescored, with the dormancy verdict, the pivot trace and the contract read behind it: ownyourmind.ai/projects/ora/.

• Every project we track carries dated, sourced verdicts across five scoring dimensions: ownyourmind.ai/resources/integrity-checks/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds GLM, VIRTUAL and VVV.

Between 4 and 7 September, four of the networks we track published a number describing how much inference they serve. Phala said 38.86 billion billed input and output tokens over 24 hours. io.net said 8 billion tokens in a single day. IoTeX said 2 billion tokens over seven days across 45 models. Heurist said its inference router is the top merchant on XRPL’s x402 network by transaction volume.

Four windows, four denominators, and no two of those figures can be put in the same column. Working out which of them anyone outside the company can check turned out to be the interesting part.

What shipped this week

• peaq put its machine economy on a subscription. Economics 2.0 went live on mainnet on 1 September 2026: a machine activates on a USD-priced tier and bonds PEAQ at the oracle rate, with the bond counted as circulating supply. Pricing machine access in dollars while settling in the token is what turns a DePIN network into something a fleet operator can put in a budget. peaq’s own documentation records the burn address for lapsed bonds as unset, so the half of a decayed bond advertised as burned currently reduces nothing. peaq tokenomics docs, read 5 September 2026

• Allora’s forecasts got a machine-shaped front door. peaq wired Allora in as a native peaqOS adapter on 3 September 2026, so a robot can pull a decentralised price or weather forecast with a keyless chain read and no account anywhere. A forecast a machine can fetch without signing up for anything is a different product from one a trader reads on a screen. peaq’s announcement says machines sell forecasts back “autonomously, on demand, without human intervention”, while the adapter page says peaqOS holds no keys and signs nothing, so the earning half still runs on an operator’s own wallet. robotic.sh adapter page, read 5 September 2026

• Heurist put its tool catalogue behind agent payments. On 5 September 2026 Heurist Mesh tools became callable and payable through x402 on XRPL, priced in RLUSD, covering token trends, DeFi metrics and NFT holdings. An agent paying per call in a stablecoin is the shape most of this sector says it is building towards, and the settlement asset here is a dollar token rather than HEU. @heurist_ai, 5 September 2026

• Coinbase closed the door on ATH derivatives. Coinbase announced the removal on 20 August 2026 and nine perpetual futures contracts stopped trading around 13:00 UTC on 3 September, Aethir’s among them, with open positions settled automatically and spot untouched. Derivatives are where a token gets priced by people who never intend to hold it, so losing the venue narrows who can take a view on ATH at all. Coinbase gives a routine product review as the reason and suspends IOTX spot on 23 September under the same kind of review, which reads as a shelf being tidied rather than a verdict on Aethir. @CoinbaseMarkets, 20 August 2026

The one number anyone can check

OpenRouter is a commercial router. You send it a prompt, it picks a provider, it bills you in dollars. It also publishes a daily per-model token series on every provider page, which makes it the only place in decentralised inference where a throughput figure is settled by somebody other than the company quoting it.

It lists 106 providers. Five are networks we cover: Venice, Phala, AkashML, Chutes (Bittensor’s subnet 64, operated by Rayon Labs) and io.net. Prime Intellect is registered as a provider with no measurable traffic. Here is the week of 1 to 7 September, read off OpenRouter’s own series on 8 September:

• Venice 373.1B tokens (53.3B a day)

• Phala 301.3B (43.1B a day)

• AkashML 166.6B (23.8B a day)

• Chutes 72.6B (10.4B a day)

• io.net 42.3B (6.0B a day)

Three things the table settles

io.net’s number is accurate and it’s their best day. OpenRouter shows 8.02 billion tokens on 6 September, against a claim of 8 billion in a single day posted the following morning. Their seven-day average is 6.0 billion. Quoting your peak is ordinary marketing, and io.net is also the only one of the four to name a source anyone could go and read.

Phala’s own figure is smaller than one router’s traffic to it. Phala’s 24-hour window closed at 14:49 UTC on 4 September. OpenRouter routed 57.4 billion tokens to Phala on 3 September and 54.9 billion on 4 September, against Phala’s 38.86 billion.

The two use different words, “billed” and “processed”, and the gap runs the unusual way round for this sector, with the self-report as the conservative number. Phala’s provider entry points its privacy policy at redpill.ai, a separate inference brand, which hints that the two figures have different scopes without explaining the difference.

Two of the four networks aren’t on OpenRouter at all. Neither IoTeX nor Heurist appears among the 106, so neither published number can be checked by anyone outside those two companies. That’s no accusation, and it’s the whole difference between a figure and a measurement.

Whose tokens were they

IoTeX’s 2 billion came through QuickSilver Pro, and IoTeX said so in the same post: 45 models, “from OpenAI, Anthropic, Goo[gle]”. QuickSilver Pro’s pricing page is operated by MachineFi Inc, IoTeX’s operating company, and lists thirteen model providers: OpenAI, Anthropic and Google among them. The page names no chain and no token anywhere on it.

So the tokens are there and a closed lab produced almost all of them. Reselling frontier models through one API key is a reasonable business and a poor measure of decentralised compute, and the two get quoted as though they were the same quantity.

What the tokens are worth

Priced at each provider’s own listed OpenRouter rates, that week converts to a gross annualised band:

• Phala $12.0M to $46.6M

• Venice $5.4M to $20.4M

• AkashML $2.3M to $9.7M

• Chutes $1.6M to $9.7M

• io.net $1.6M to $5.6M

The bands are wide because OpenRouter’s series carries no prompt/completion split and the two rates differ by up to several times, so each floor prices every token as input and each ceiling prices every token as output. It’s the same method our Bittensor demand roll-up already uses. The truth sits between the two ends, and the floor is the one to quote if you want a single number, because the ceiling assumes every token billed at the dearer output rate.

Two caveats sit on those figures. This is gross revenue to the router before OpenRouter’s cut and before the GPU supply gets paid, so the network keeps a fraction of it. And it’s one channel: every one of these five sells directly as well, and none of them publishes that side.

Then the part that matters for a token holder. On a single DeepSeek model I counted 28 providers competing, Phala and Venice among them, alongside the big clouds and the frontier labs’ own endpoints. The buyer picks on price and latency. OpenRouter settles in dollars with a company, and PHA, VVV, AKT and TAO are nowhere in the transaction.

What to do with a token count

Three things worth carrying to anyone’s throughput number, ours included:

• A token count needs a window and a producer. Two billion over seven days and 38.86 billion over 24 hours differ by more than a hundred times per day, and one of them counts models built by somebody else.

• Ask who else can see it. A figure only the issuer can compute is a claim. The same figure on a third party’s page is a measurement, and it stays one even when the third party is centralised.

• Volume isn’t revenue, and revenue isn’t the token. Every step between the three loses most of the number, and the last step frequently loses all of it.

Biggest movers this week

7-day change as of 7 September 2026. Twenty-nine of the forty tokens we track closed up.

Gainers:

• FLock.io (FLOCK) +67.3%

• peaq (PEAQ) +33.8%

• NuNet (NTX) +33.4%

• NEAR Protocol (NEAR) +28.0%

• Fetch.ai / ASI Alliance (FET) +20.3%

Fallers:

• Ora Protocol (ORA) -30.2%

• Giza (GIZA) -15.9%

• Heurist (HEU) -15.6%

• ElizaOS (ELIZAOS) -11.2%

• Openmind (ROBO) -8.3%

See all movers: ownyourmind.ai/projects/movers/

What we verified this week

Nine scores moved. Five findings are worth your time.

• IoTeX hasn’t held a governance vote since October 2024. Our evidence called the IIP process functional and accelerating, and cited IIP-42 as passing on 60 voters. The Snapshot space has put nothing to a vote since IIP-40 on 28 October 2024, and IIP-42 reads Status: Draft in IoTeX’s own repository. Governance dropped from 10/20 to 8/20.

• The half of our old IoTeX claim that survived is the interesting half. IIP-42 proposed halving the block time, and the chain does now produce a block about every 2.5 seconds, measured over a fortnight on IoTeX’s own RPC. A repository status tracks the paper and never the chain, so a Draft document can accurately describe something that already shipped.

• A settled lawsuit that was never settled. Ten surfaces across our Ocean Protocol coverage described the Fetch.ai litigation as resolved with 286 million FET returned. The docket carries no termination, briefing ran to 26 June 2026, the 286 million is an alleged conversion rather than a recovery, and the plaintiffs are a putative class rather than Fetch.ai.

• Heurist’s chain is starker than our own prose. We had reasoned from the allocation table that a paused mining programme functionally re-routes half of supply to foundation discretion. On-chain it’s more direct: one Safe holds 66.20% of supply, and eighty-two addresses hold the token at all.

• peaq folded two labelled reserves into the team-controlled one. The Community and Security Reserves each sent their full balance to the Ecosystem and Treasury Reserve on 28 August 2026 and are now empty, with a further tranche leaving on 1 September against a stated purpose of “to be confirmed”. The total is unchanged; about a fifth of supply now sits behind a single label.

Behind those, every project we track now carries a dated, sourced verdict on which exchanges actually trade its token, checked at each venue’s own API rather than at an aggregator. Twenty-two of thirty-six needed correcting, in both directions. Giza’s five named venues have all gone; Kraken lists NOS, which our evidence had denied.

See the full integrity board: ownyourmind.ai/resources/integrity-checks/

On the watchlist

• Whether Venice’s seventh emission cut lands where the last six did. VVV emissions dropped from 3M to 2.5M a year on 1 September, the seventh reduction in about two years, against a stated goal of net deflation. Venice is also the largest of the five on OpenRouter and the only one that published no volume figure this week.

• Whether peaq switches its burn address on. Economics 2.0 is the machine-economy thesis written into contracts, and the one part of it that would make PEAQ scarcer currently points at an address the project’s own docs record as unset.

• Whether any of these networks routes inference revenue through its token. Five of them sell on OpenRouter and get paid in dollars. Akash’s burn mechanism is the only one of the five designed to convert compute demand into token demand, and its realised burns have so far offset a fraction of issuance.

From the site

• The IoTeX episode went out on the podcast on 6 September, recorded before the governance finding above, which is a good illustration of why the written review carries the score and the audio doesn’t: ownyourmind.ai/projects/iotex/.

• Every project we track now carries dated, sourced verdicts across five scoring dimensions: ownyourmind.ai/resources/integrity-checks/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds GRASS and NEAR.

Aethir announced the two things we mark its token down for. On 24 August 2026 it launched ACCELERATE, a buildout of ten data-centre sites in the US and Europe, and said an updated tokenomics policy will add a burn mechanism and a variable platform fee. Our review scores ATH’s value accrual low for having no fee distribution and no buy-and-burn, so on paper this is the fix.

Then we read who is building it. The partner is a Nasdaq-listed company whose treasury is ATH, which until nine months ago filed accounts as a cancer-diagnostics business, and in which the Aethir Foundation says it holds shares.

What shipped this week

• Coinbase started trading Grass. Full GRASS-USD spot trading opened on 26 August 2026, five days after Coinbase added the token to its listing roadmap. It’s the deepest single venue a bandwidth-sharing DePIN token is likely to reach, on a project we already mark down for opaque foundation control of revenue. One correction to how it’s being written up: Kraken is already on Grass’s exchange list and is US-regulated, so Coinbase widens American access rather than opening it.

• Coinbase is closing IOTX in the same fortnight. From 23 September 2026 the token can no longer be bought or sold there, though deposits and withdrawals stay open. IoTeX’s review cites Coinbase by name as evidence of institutional access, so losing it is a scoring question. It isn’t a verdict on IoTeX, either: the same listing-standards review named Badger DAO and Storj five days later.

• TAO can now be swapped on Base, and almost none of it is there. Chainlink’s CCIP standard carries the token to Base, where an Aerodrome pool trades it against USDC, reported on 29 August 2026. A second venue matters for a token whose demand has been bounded by its own chain. We read the bridge contract rather than the announcement: 2,128 TAO have crossed, about two hundredths of one per cent of circulating supply.

The builder and the holder

ACCELERATE is the biggest capital commitment in Aethir’s history. Access to ten sites in the US and Europe, up to 20 megawatts, NVIDIA B300 and GB300 clusters of 64 to 256 nodes each, targeting up to $700 million in contract value by the end of 2026 and over $2 billion once all ten are built. The announcement runs alongside an updated tokenomics policy that Aethir says “will include” a burn mechanism and a variable platform fee.

Hold that verb. A burn that will be included is a plan, and our review’s line that there is no fee distribution and no buy-and-burn stays accurate until a contract says otherwise. Aethir’s whitepaper tells you what the team intends; only deployed code tells you what happens to your money.

The buildout partner deserves more attention than the megawatts. Aethir names Axe Build, a division of Axe Compute Inc. On EDGAR that is CIK 0001446159, Nasdaq ticker AGPU, filed under SIC 6199 and assigned to the SEC’s Crypto Assets office. The same registrant filed as Predictive Oncology Inc. until 11 December 2025, and before that as Precision Therapeutics, Skyline Medical and BioDrain Medical.

Its balance sheet is the interesting part. Axe Compute’s annual report for the year to December 2025 discloses 2.837 billion unlocked ATH plus rights to a further 3.511 billion, about 15% of Aethir’s 42 billion maximum supply in one listed company. Our own Aethir review already recorded that investment and said the relationship between the two treasury structures was unclear. It isn’t unclear any more: the largest disclosed ATH holder is now the contractor building Aethir’s data centres.

Aethir discloses the tie in its own announcement, in one line: “Aethir Foundation is an Axe Compute shareholder.” Axe Compute’s annual report, filed for a year that ended before ACCELERATE existed, records no equity in either direction, so the shareholding postdates the last audited account and nobody has yet had to put a number on it.

Meanwhile, what are customers actually paying? We index every service-fee deposit on the AETHIR_CORE contract on Arbitrum, which is the one compute network in our set whose revenue reconciles against an independent adapter. Gross fees in dollars fell about 41% between the first and second quarters of 2026, and the current run-rate sits near a quarter of its peak. Sixty-one addresses have ever paid a service fee, so what is thinning is a handful of enterprise contracts.

Now the part that makes the promised burn worth watching rather than celebrating. Counted in ATH instead of dollars, that same fee stream more than tripled year on year. Deposits track the token price at a correlation of -0.90, which is the signature of contracts priced in fiat and settled in tokens: when ATH halves, twice as many tokens cross for the same invoice.

A burn denominated in ATH therefore grows as ATH gets cheaper. It would look most impressive in exactly the quarters when the dollars behind it are smallest, and a holder reading a rising burn total as rising demand would have it backwards. The number to watch when the policy lands is the dollar value of fees, not the token count of the burn.

What would change our mind

• A deployed burn contract with an address. Aethir’s supply is 42 billion with under half circulating and compute-provider allocations still vesting, so a burn that actually runs against gross fees is the single largest thing that could move its supply and value-accrual scores.

• The dollar fee line turning up. The run-rate has fallen for three complete quarters. ACCELERATE is a bet that owning racks wins contracts that renting other people’s racks did not.

• Arm’s-length revenue. The ATH treasury company, the compute buyer and the site builder are now one Nasdaq registrant, with the Aethir Foundation on its share register. Vertical integration can be a sound commercial decision. It also makes a revenue figure much harder for an outsider to read, and Aethir’s revenue was already the strongest thing it had.

Full review, scores and the fee index: ownyourmind.ai/projects/aethir/

Biggest movers this week

7-day change as of 30 August 2026. The week split the cohort down the middle, with 20 of the 40 tokens we track closing up. Grass leads on its Coinbase listing and is a position we hold, so read it against the disclosure at the foot. Heurist is the sharpest reversal: it topped last issue’s gainers, and last issue was where we took its headline claim apart.

Gainers:

• Nillion (NIL) +26.6%

• peaq (PEAQ) +26.4%

• Grass (GRASS) +21.9%

• Cookie DAO (COOKIE) +16.1%

• Ora Protocol (ORA) +15.9%

Fallers:

• OpenServ (SERV) -21.0%

• Heurist (HEU) -19.1%

• Warden Protocol (WARD) -13.6%

• NuNet (NTX) -9.8%

• Auki (AUKI) -9.3%

See all movers: ownyourmind.ai/projects/movers/

What we verified this week

Four claim-checks in five days, and no score moved. One of them corrects this newsletter.

• We overstated peaq’s World ID shipment, in the last issue. Issue 13 said peaqOS integrated World ID, which is how three outlets reported it. All three attribute it to “a media statement”, so they are one source wearing three bylines. On robotic.sh, peaq’s own machine-services market, World ID carries the status Setup, meaning an operator has to configure it, while most of the catalogue is marked ready.

• A second token on Base is also called “Bittensor TAO”. The bridged TAO carries a live CCIP admin and a supply consistent with a 21 million cap. The impostor carries a billion tokens. Anyone reading a TAO-on-Base figure off the wrong contract would be reading a number nearly fifty times the entire real supply.

• Aethir’s buildout partner is a renamed cancer-diagnostics company. Verified against EDGAR rather than either company’s marketing, which is the point: both firms describe the relationship in press releases, and only the registrant history shows what was being filed nine months ago.

• An exchange listing was being framed as opening US access that already existed. Grass has been tradeable on a US-regulated venue for months. Coinbase deepens that access, which is worth writing down accurately, because a score built on the wrong premise is harder to correct later than one built on none.

See the full integrity board: ownyourmind.ai/resources/integrity-checks/

On the watchlist

• Whether Aethir’s tokenomics policy ships as code. The gap between “will include a burn mechanism” and a contract address is where most tokenomics announcements go to die. We will read the address, not the blog.

• Whether IoTeX’s delisting is the start of a pattern. Coinbase named three assets on one listing-standards review. If US venues keep tightening, exchange access stops being a stable input to a Returns score and becomes something to re-check every quarter.

• Whether TAO’s Base float grows. Two thousand tokens is a proof of plumbing. If that number climbs into a meaningful share of circulating supply, the liquidity picture changes and the score follows it.

From the site

• The movers page now does windows and exports. Seven-day, one-month, six-month and one-year toggles on the same basis as the dashboard trend, plus a one-click PNG of the snapshot card: ownyourmind.ai/projects/movers/.

• Every tool now sits behind one hub, including the Morpheus model, the tokenomics simulator and the GPU income tracker: ownyourmind.ai/tools/.

• The site stopped calling out to Google for fonts. They are self-hosted now, which fixes image export in browsers that block third-party font requests and removes one third party from every page load.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds GRASS and MOR.

Value accrual is a narrow question. Does using a network give its token a claim on anything, and if so, how much? Almost every published answer for Morpheus comes from the project’s own documents. Over the past fortnight we rebuilt the whole economy from the deployed contracts instead, checked it against eight months of chain history, and put it on the site as something you can drive yourself.

The answer splits in two, and conflating the halves is how most MOR analysis goes wrong. Usage pays the token almost nothing today. What MOR has instead is a set of unusually tight supply mechanics, and they’re doing the work people credit to demand.

What shipped this week

• Nous Research turned its agent into a team. Bot Mode shipped in Hermes Desktop on 18 August 2026, replacing one-off sessions with named, persistent bots, each with its own model, memory and skills, that hand work to each other by @mention. Build a specialist once and it’s still there next month, which is the gap between a chat window and a colleague. Nous is upfront that every bot still has full system access and needs the same supervision as before. MarkTechPost, 17 August 2026

• The same agent now reads a skill before installing it. From 19 August 2026 Hermes can run NVIDIA’s open-source SkillEvaluator over an incoming skill, surfacing personal data, leaked secrets, Unicode smuggling and licensing problems line by line. An agent skill is code that inherits your machine’s permissions, and hardly anyone reads one first. The scan is optional and advisory, so it informs the decision without blocking the install. NVIDIA developer blog, 19 August 2026

• A robot can now check you are human without learning who you are. peaqOS integrated World Network’s World ID on 21 August 2026, so a machine receives a zero-knowledge proof of personhood and keeps an auditable record, with no name or face collected. Machine-to-human interaction has assumed that verifying identity means collecting it, and this pulls the two apart. peaq’s own structure is untouched by it: still a Polkadot parachain, and a 45% insider allocation still concentrates the on-chain vote. Crypto Briefing, 21 August 2026

• Heurist put stablecoin rails under agent usage. From 17 August 2026 its AgentCore handles pay-per-use billing, paywalled content access and spending caps, all settled in USDC. An agent that can pay for what it consumes stops needing a human to top up an account, which is the unglamorous piece the agent economy has been short of. HEU sits outside the settlement path, and the claim Heurist attached to the launch is this issue’s verification story below. Heurist, 17 August 2026

What the model says about value accrual to MOR

Morpheus mints MOR on a fixed schedule and splits it between five pillars: capital, compute, code, builders and a protection fund. We rebuilt each one from the contracts and checked it against the chain. The emission machinery is sound, so the question worth asking is what happens to the MOR after it’s minted.

Almost nobody pays for inference in MOR. Our index of every closed session shows 63.2 billion tokens served since launch. Users paid for 48,138 of them. The rest came out of emissions, because paying a provider directly buys you no more session time than staking does.

The protocol does buy MOR on the open market, but that money is staking yield on capital deposited with it, and it would carry on unchanged if usage fell to zero. So a MOR holder receives no income from anyone using the network. If that’s what you thought you were buying, it isn’t there yet.

The supply side is a different story. We mapped every address across the three chains that holds MOR for a structural reason. Around four fifths of the supply falls into one of four buckets:

• Held by the protocol itself, 41%. Treasury Safes plus the 16-year emission lock.

• Locked in contracts by users, 30%. Builder-subnet stake, provider collateral, session stake.

• Liquidity-pool inventory, 3.5%.

• Burned, 3%.

That leaves a 22% residual: ordinary holders, exchanges, and wallets nobody has identified. Read it as a ceiling on how much MOR could be sold, not a measure of what’s liquid. Circulating supply counts the treasuries and the locked stake, and comes out four times larger.

Two things eat into it. The buyback takes about 3.4% of daily issuance off the market, burning half and locking the rest away for sixteen years. And a provider can earn no more MOR than it has staked, over a rolling year from its first claim, so a provider growing its business is a provider locking up more.

Which brings us to the only way using Morpheus reaches the token. You can’t draw compute without staking MOR, and you can’t stake what you don’t own, so a user has to buy first, out of that residual. No fee is paid to anyone; the token just gets harder to find.

At the model’s assumed growth the residual is exhausted by August 2027. Issuance itself falls in a straight line to zero in 2040, against a 42 million cap.

The caveat is a big one. All of it depends on people carrying on staking fresh MOR for compute they aren’t paying for, and our eight months of history covers a bootstrap, a spell of contract bugs and one operator moving its own treasury around. The model marks that dial an assumption rather than inventing a growth rate.

Three things to check if you hold MOR

• Which chain your builder stake sits on. The Arbitrum deployment’s reward share has been zero since launch, so roughly one MOR in six of all builder stake earns nothing across 124 subnets, listed on the dashboard beside the paying ones with no visible difference.

• What the lock multiplier actually buys. The advertised 10.7x sets your share against other stakers, and the pool’s blend sits near 2x. On 25 August that was 5.05% a year unlocked against 10.35% at the blend.

• Where the buyback’s MOR ends up. Our own pages carried a 75/25 buy-and-pair split from the project’s improvement proposal. On chain it’s 50/50 burn and lock, and the liquidity leg last received anything in January 2025. A whitepaper tells you what a team intended; only the contract tells you what happens to your money.

Drive the model yourself: measured inputs marked measured, the two guesses labelled as guesses, every scenario encoded in the URL.

Biggest movers this week

7-day change as of 25 August 2026. Another broadly green week, 33 of the 40 tokens we track closed up, and compute led it with a mean of about +22%. The week’s biggest gain went to Heurist, which is also the project whose headline claim this issue takes apart below. Virtuals sits fourth among the gainers and is a position we hold, so read that one against the disclosure at the foot.

Gainers:

• Heurist (HEU) +54.9%

• OriginTrail (TRAC) +43.4%

• Fetch.ai / ASI Alliance (FET) +42.3%

• Virtuals Protocol (VIRTUAL) +38.3%

• Aethir (ATH) +35.7%

Fallers:

• Nillion (NIL) -12.7%

• Allora Network (ALLO) -11.2%

• Giza (GIZA) -7.0%

• Warden Protocol (WARD) -5.2%

• ElizaOS (ELIZAOS) -1.7%

See all movers: ownyourmind.ai/projects/movers/

What we verified this week

Thirteen independent claim-checks since the last issue, and no score moved. That is the ordinary result when the checking runs continuously instead of only at review time.

• An announcement rode on a name collision. Heurist paired its payments launch with the claim that OpenAI had adopted “the exact agentic payment technology that powers every transaction on Heurist Finance”. The cited guide, published 13 August 2026, pairs OpenAI’s Agents SDK with Amazon Bedrock AgentCore Payments, an AWS product that went generally available on 18 August 2026. The shared component is x402, an open standard anyone can build on. Opening the cited source, instead of the post quoting it, is the whole check.

• We corrected our own diagram before anyone else did. The value-loop graphic on the Morpheus project page described a liquidity mechanism that stopped operating in January 2025, while the deep-dive two clicks away had it right. Fixed, and the footer now dates the change instead of carrying an acronym the diagram no longer uses.

• An aggregator series was flapping, so we stopped quoting it. DeFiLlama’s NEAR chain TVL swung by roughly 40% a day across 15 to 19 August 2026. We traced it to a single lending market whose supplied leg drops in and out of the adapter. The freshness verdict is paused instead of published, because routing a spurious drift every few days would be worse than carrying no figure.

• Nine reviews lost hardcoded valuations. A price-agnostic sweep pulled drifting market caps and dollar volumes out of the evidence rows on Auki, Bittensor, Golem, Heurist, NEAR, NuNet, OpenServ, Render and Sentient, and restated each one structurally. A figure that goes stale between reviews was never carrying the argument anyway.

See the full integrity board: ownyourmind.ai/resources/integrity-checks/

On the watchlist

• Whether Arbitrum’s builder share ever gets set. It’s a single administrative call that has never been made on that chain. Making it would start paying a stake that has earned nothing since deployment, and the size of the backdated question is the reason it may stay unmade.

• Whether committed access stake keeps arriving. Provider pay follows a stipend formula the contract applies to staked access, so the contract damps its own draw and the pool spends down without ever emptying. That leaves one behavioural question the code can’t answer: whether new access stake keeps coming. It’s the one input our model refuses to put a measured growth rate on.

• Whether the advisory scan becomes a blocking one. Nous shipped skill scanning as advice, with no power to refuse. Agent skills run with real system permissions, so the moment a scan starts refusing installs is the moment it becomes a safety mechanism rather than a label.

From the site

• The Morpheus model has its own page, with the emission schedule, the access stock and the free-float line all driveable, and each scenario linkable by URL: ownyourmind.ai/tools/morpheus-model/. The reasoning behind it sits in the tokenomics deep dive.

• How Morpheus builder subnets work gained the Arbitrum section described above, with the contract figures bound live rather than typed: ownyourmind.ai/tokenomics/how-morpheus-builder-subnets-work/.

• Staking for inference compares the three routes that let you stake a token and draw inference against it, across Morpheus, NEAR and Venice: ownyourmind.ai/tokenomics/staking-for-inference-morpheus-near-venice/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds AKT, GLM, GRASS, MOR, NEAR, RENDER, VIRTUAL and VVV.

A stated token supply is one of the easiest things in crypto to check, and one of the least often checked. Over the past fortnight we read the deployed contracts behind most of the tokens we cover. The stated number almost always matched. What varied, and varied a lot, was whether anything in the code stops that number changing.

What shipped this week

• NEAR handed its privacy check to Intel. Since 12 August 2026, every confidential inference request gets an Intel-signed token certifying the prompt ran inside a sealed chip and stayed encrypted. Confidential inference is worth whatever its proof is worth, and this puts the proof with a party that has no stake in the answer. Phala has done the same since March 2026 and SCRT Labs since May, so a shared standard is spreading. Crypto Briefing, 12 August 2026

• Allora now promotes and relegates its forecasters automatically. From 13 August 2026 a worker earns its place on the live network by passing seven statistical tests on a mirrored practice network, and drops back the moment its accuracy slips. Deciding who gets to supply a network is power, and Allora has handed that call to published numbers. Allora blog, 13 August 2026

• Akash turned the idle gaming PC into a rentable server. Homenode reached open beta on 11 August 2026, with Windows and Linux installers and support for consumer 3090s, 4090s and 5090s. A compute marketplace is only as decentralised as the number of people supplying it, and that is Akash’s weak spot: Messari put its average active providers at 58 in Q1 2026. homenode.akash.network

• Bittensor put a price on subnet takeovers. From 14 August 2026, spec 447 makes a single account hold 18% of a subnet’s locked stake before it can even attempt one. Whoever controls a subnet controls where its emissions go, which is the money its miners and validators are working for. This closes one route to that, and the single sudo key described below is another that stays open. subtensor PR #3083, merged 14 August 2026

The revenue number we stopped citing

Bittensor shipped three mechanism changes in eight days. Mentat Lend opened on 6 August as the network’s first native credit market, letting alpha holders borrow TAO against subnet positions that stay staked and keep earning. The takeover gate above, which Bittensor calls Conviction, landed on the 14th. Root Reborn, which we covered last issue, landed on the 4th. All three move value around more efficiently inside the network.

The thing that would change the investment case is money arriving from outside it. Since March our review leaned on Pine Analytics’ bear case for that number, the way most Bittensor coverage still does. This fortnight we stopped citing it and measured the demand leg ourselves.

What can be settled independently is thinner than the cited version. It amounts to one subnet’s inference traffic on a public router, priced at that subnet’s own listed rates, against the TAO the network mints each day derived from the chain’s own issuance delta. At the generous end, counting every self-report on top, external demand covers 6.4% of the emission budget. Our previous framing was wrong in both directions: we had carried “under 5% of emissions”, and the budget behind that ratio had been priced off a TAO price roughly a third above where the token actually traded.

Grayscale’s Q2 rebalancing, announced 5 August, cuts the other way on sentiment. The fund sold NEAR and spread the proceeds across its remaining components, leaving NEAR the largest weight at 31.35% and TAO second at 29.15%. Institutional demand for exposure is a signal about who wants the token. It says nothing about who pays the network.

Governance is the part that didn’t improve. Read the deployed chain code and there’s no council, senate or voting body in it at all: privileged operations run through a single administrator key. The takeover gate above helps, but it’s only half a mechanism, and the voting half is switched off across the network. We grade every project on decentralisation and on the investment case; Bittensor held a C on both this fortnight, with the revenue call due at the September review.

Your prompts are turning into evidence

The strongest argument for private AI came from a courtroom rather than a pitch deck. In the New York Times copyright case against OpenAI, a federal magistrate ordered the company to produce a sample of 20 million ChatGPT conversation logs, and the district judge upheld it over OpenAI’s objection that handing them over would invade user privacy. The plaintiffs had asked for 120 million. The sample is anonymised, and the people whose chats became evidence were never party to the case and were never asked.

Corporations worked this out earlier. Samsung restricted internal ChatGPT use back in 2023, after engineers pasted proprietary semiconductor source code into it three times in under a month. Most firms have no such policy and quietly hope their staff are careful.

The takeaway is about where AI runs. “Trust us, it’s private” is a policy, and a policy can be overridden by a subpoena. Hardware-attested inference, where a chip proves your prompt ran inside a sealed enclave and was never logged, is the version that survives one. That is why NEAR handing the attestation check to Intel this week matters more than the marketing around it: the claim became something a third party can check.

Biggest movers this week

7-day change as of 19 August 2026. Ocean led the upside and Ora led the other way, about fifty points apart end to end. Venice sits second among the gainers and is a position we hold, so read that one against the disclosure at the foot.

Gainers:

• Ocean Protocol (OCEAN) +22.8%

• Venice (VVV) +16.7%

• Nosana (NOS) +5.1%

• Virtuals Protocol (VIRTUAL) +4.9%

• Theta Network (THETA) +4.4%

Fallers:

• Ora Protocol (ORA) -27.5%

• Olas (OLAS) -26.2%

• Sentient (SENT) -13.9%

• Heurist (HEU) -13.5%

• Cookie DAO (COOKIE) -12.4%

See all movers: ownyourmind.ai/projects/movers/

What we verified this week

The verification board took 113 independent claim-checks since the last issue, most of them reading deployed contract state rather than documentation. No score moved, which is the point of doing it this way: the checks either confirm what we published or correct it.

• A hard cap and an enforced cap are different things. Across the sweep, the stated supply nearly always matched what the chain reports. Enforcement is where projects separate. Grass’s mint authority is a live single-keypair wallet. Fetch.ai’s token exposes an uncapped mint behind one un-renounced multisig. Giza’s sits behind a small multisig with no timelock at all.

• Several caps are closed for good, and we said so. Nosana’s mint authority is null, so no more can ever be minted. Olas’ cap is arithmetic inside the contract, with issued supply and the remaining allowance summing exactly to the stated total. Golem’s two token legs also sum exactly, which makes its migration supply-conserving rather than issuance.

• We stopped citing Bittensor’s revenue problem and measured it. The centrepiece above is the result. It corrected our own published framing as much as the project’s story.

• A design document isn’t a deployed contract. Five Morpheus access-mechanism claims turned out to describe a published model rather than the code running on-chain. All five were rewritten against the deployed contract, and the gap between the two is now its own record.

• NEAR’s governance and its config disagree. Its DAO approved removing the developer gas rebate in June 2026, and the live protocol config still returns the rebate. We also enumerated the validator set on-chain and found it materially smaller than the figure we had carried, because the old one counted staking pools.

See the full integrity board: ownyourmind.ai/resources/integrity-checks/

On the watchlist

• Whether any of the open mint paths close. Renouncing a mint authority or moving it behind a timelock is a cheap, visible action. The projects that take it will be easy to tell apart from the ones that explain why they don’t need to.

• Bittensor’s Conviction voting power. The takeover half of that mechanism shipped this week, but the voting half is still switched off across the network, with totals reading zero. Switching it on would create an on-chain governance route where the runtime currently has none.

• NEAR’s gas rebate gap. A governance decision that the protocol config hasn’t implemented is either a pending upgrade or a dead letter. Which one it turns out to be says a lot about what its DAO actually controls.

From the site

• Staking for inference went up this month, comparing the three routes that let you stake a token and draw inference from it, across Morpheus, NEAR and Venice: ownyourmind.ai/tokenomics/staking-for-inference-morpheus-near-venice/.

• The Chutes piece was rebuilt this week on our own measured figures instead of the cited bear case, which is the working behind this issue’s centrepiece: ownyourmind.ai/tokenomics/chutes-bittensor-revenue-machine/.

• Bittensor subnet revenue was refreshed alongside it, tracking where inside the network the money actually lands: ownyourmind.ai/tokenomics/bittensor-subnets-where-the-revenue-is/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds AKT, GLM, GRASS, MOR, NEAR, VIRTUAL and VVV.

Three protocol releases landed in the same week, and a fourth project wound its token down entirely. That makes it a good week to talk about a habit worth having. The distance between what a release announcement claims and what is actually switched on is where most of the useful information sits. Bittensor gave us the clean case, AntSeed gave us the mirror image, and Gensyn is still giving us the long-running one. Read the patch notes, then read the post.

What shipped this week

• Bittensor shipped Root Reborn (runtime v441) on 4 August 2026. Root-stake yield in subnet alpha used to be sold to TAO mechanically every block. It now accrues in per-validator escrowed basket funds and is realised only when a root staker claims, which removes a constant, price-insensitive sell stream on subnet alpha. Opentensor, Root Reborn docs

• Nous Research shipped Hermes Agent v0.20.0 on 3 August 2026, the largest release in the project’s history by its own count: roughly 3,650 commits and 1,400 merged pull requests from 650-plus contributors since v0.19.0. The headline additions are real-time conversational voice with barge-in, agent-to-agent messaging on A2A v1.0, and signed outbound webhooks. That moves Hermes from a text assistant toward a multimodal agent platform. GitHub release notes, 3 August 2026

• AntSeed announced a verifier plugin SDK on 2 August 2026, though the code merged on 21 July. Buyers can now demand cryptographic proof, starting with Intel TDX attestation, that a peer is running the advertised model on the advertised hardware. Policy is best-effort by default, with a hard-block flag for buyers who want it enforced. AntSeed PR #713, merged 21 July 2026

• ElizaOS founder Shaw Walters declared the ELIZAOS token dead on 4 August 2026 and confirmed the foundation is closing, after its treasury went to settling the Burwick Law class action. Per CoinDesk he ruled out buybacks, supply measures and any replacement token, while committing to keep building the MIT-licensed framework. CoinDesk, 5 August 2026

The gap between the announcement and the switch

Root Reborn is the worked example. The Opentensor Foundation marketed a competitive allocation layer where validators assemble subnet baskets and compete on returns, and that framing carried most of the coverage. Checking the runtime against the docs, the curation layer is gated off at launch. Every basket fund currently runs a null accumulate-in-place strategy until a later upgrade enables it, so the thing validators are supposed to compete on isn’t live yet.

The part that did ship still matters, and probably matters more. A permanent block-by-block sell stream on subnet alpha has been replaced by yield that sits in escrow until a staker claims it. That is a structural change to sell pressure across every subnet, and it landed on day one. The full Bittensor review carries the mechanics and what it does to the emissions model.

AntSeed ran the same story backwards. The verifier SDK arrived as an ordinary pull request on 21 July and only got its announcement on 2 August, so the code was in the tree for nearly two weeks before anyone framed it. It’s also the more consequential shipment of the two. Buyers on a compute marketplace can now hard-block a peer that can’t attest to running the advertised model, which is the first piece of the validation system that gates AntSeed’s locked provider emissions.

Gensyn is the standing version of the same question. The Delphi Agent Arena competition opened on the L2 this week and the prediction market is live, while the training-network mainnet, the actual product, remains unfinished. The token has traded since April 2026. Token utility, value accrual and revenue sustainability all sit behind a launch that hasn’t happened.

None of this makes the announcements dishonest. It makes them forward-looking, which is a different thing to read for. When a release lands, the question that pays is simple: which parts of this are switched on today?

Biggest movers this week

7-day change as of 11 August 2026. ElizaOS halved in the week its founder told holders to sell, which is the rare case of the tape and a public statement agreeing outright. Giza went the other way in a violent round trip, leading the fallers last issue and leading the gainers in this one.

Gainers:

• Giza (GIZA) +84.9%

• Cookie DAO (COOKIE) +59.7%

• Allora Network (ALLO) +21.0%

• IoTeX (IOTX) +13.5%

• Nillion (NIL) +11.4%

Fallers:

• ElizaOS (ELIZAOS) -50.5%

• Heurist (HEU) -39.3%

• OpenServ (SERV) -24.3%

• NEAR Protocol (NEAR) -13.4%

• Akash Network (AKT) -12.1%

See all movers: ownyourmind.ai/projects/movers/

What we verified this week

Twenty independent claim-checks landed across the projects we cover since the last issue, most of them from a cold-start grounding run that reads deployed contract state rather than documentation.

• Stated supply caps mostly survive contact with the contract. Sentient, Heurist, OriginTrail and IoTeX each had their cap checked against deployed code. All four matched, with OriginTrail’s mint permanently disabled on-chain and Heurist’s cap enforced inside the mint function itself. Sentient carries a caveat: the proxy remains upgradeable.

• Holder concentration is where the published numbers bend. Sahara AI and Ora Protocol both came out overstated on distribution fairness once infrastructure addresses were stripped from the top-10 figures. OriginTrail and OpenServ matched their claims, because roughly half of their apparent concentration is staking and bridge contracts rather than holders.

• NEAR Intents throughput re-checked first-hand. Cumulative volume through Intents is up by about a quarter since our June figure, and the fee run-rate was refreshed across the review from the source rather than carried forward.

• A model privacy tier corrected on our own page. Our Venice review had Kimi K3 on the wrong tier; the live model API says private, so the page was corrected. Venice Returns moved +1 on separately strengthened on-chain burn.

• Root Reborn checked against the runtime docs. That check is where this issue’s centrepiece came from, and it produced a Data row rather than a score move.

See the full integrity board: ownyourmind.ai/resources/integrity-checks/

On the watchlist

• Whether Bittensor switches basket curation on. The competitive layer is the reason Root Reborn was marketed as a rewrite. Until curation is enabled, root staking has changed when yield is realised rather than who allocates it.

• Gensyn’s training-network mainnet. The prediction market and the hackathon are live, the compute product isn’t. That launch is the gate on every part of the token thesis.

• AntSeed’s locked provider emissions. The verifier SDK is the tooling half of the validation system that gates seller emissions. Slashing parameters, an appeals path and making those emissions claimable are the half still outstanding.

From the site

• The Bittensor subnet economics piece was rewritten this month to the current price-based emissions model, which is the backdrop Root Reborn now sits on: ownyourmind.ai/tokenomics/bittensor-dtao-subnet-economics/.

• The ElizaOS versus Virtuals comparison was refreshed this month and reads differently now that one side of it has wound its token down: ownyourmind.ai/tokenomics/elizaos-vs-virtuals/.

• The integrity board is the running record behind the verification section above, one row per claim we have checked against an independent source: ownyourmind.ai/resources/integrity-checks/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds AKT and NEAR.

Here’s the tokenomics read worth your time this week. Venice’s DIEM mechanism lets you lock staked VVV to mint DIEM, and each DIEM pays out $1 a day in API inference credit. The catch is the mint curve: the closer supply sits to its target, the more sVVV you must lock per DIEM. This week two things moved the cost the same direction, and the second one lands on Monday.

What shipped this week

• peaq launched Economics 2.0 on 30 July 2026, folding its old six-pool model into four primitives: machine activation, trust validators, a machine pool, and a unified currency flow. Public staking now sits behind trust validators alone, and machine exit permanently destroys half the bonded position. Both systems run in parallel through a transition window. peaq blog, 30 July 2026

• NEAR brought Confidential Intents into Meteor Wallet, extending restricted-visibility cross-chain execution from the near.com app to a third-party wallet. The confidentiality rests on TEE hardware, and NEAR exposes a public attestation endpoint so the claim can be checked on-chain by anyone. NEAR Protocol, 28 July 2026

• Phala said it brought Kimi K3, Moonshot AI’s 2.8-trillion-parameter frontier model, live for confidential inference inside GPU TEEs. The model runs inside a hardware-attested enclave, so a user can verify the execution environment before trusting it with a prompt. Phala Network, 29 July 2026

• Nous Research shipped Hermes Agent support for Buzz, Block’s open-source self-hostable Nostr workspace, with three integration paths spanning a desktop runtime and a relay bridge. The link runs over Nostr, so the agent plugs into a workspace the user can host themselves. MarkTechPost, 31 July 2026

The mechanism: why the DIEM mint eases this week

DIEM is Venice’s staking-into-compute loop. You lock staked VVV in escrow to mint DIEM, stake the DIEM for $1 a day of API credit against any model on Venice, and burn the DIEM to reclaim your sVVV. The mint rate follows an exponential curve, so as supply climbs toward the target, the sVVV you must post per DIEM rises sharply, and the last stretch before the target is where it bites hardest. The full mechanism breakdown has the formula and the maths.

Two things eased that cost this week: supply net-burned further down the curve, and Venice stepped the target up. Supply now sits near 36,800 DIEM, about 96% of the raised 38,500 target, and the lock has fallen to roughly 518 sVVV per DIEM. That is down from about 601 a week ago, and from an April 2026 peak near 685.

The target keeps climbing to 40,000 in four steps of 500, on 3 August, 17 August, 31 August and 14 September 2026. The first landed on Monday, and each further step slides the supply ratio down and eases the lock again.

The DIEM mint-rate curve: sVVV locked per DIEM against DIEM supply, with the current position marked well left of the 38,500 target.

The lock stays shallow until supply nears the target, then goes near-vertical. Today’s supply sits in the flat zone, and stepping the target right keeps it there. Live version: ownyourmind.ai/projects/venice/diem/.

For anyone weighing the mint path, that’s the one actionable read: the sVVV cost of entry has dropped, with three more target steps to come through 14 September. The mechanism is doing what it’s designed to do, correcting the curve when supply and target drift apart.

The structural question sits one layer down. A net burn releases locked VVV back into circulation, which loosens the contractionary pressure the flywheel depends on. Venice’s counterweight went live on 17 July 2026: $5 of every $100 in credit purchases now buys and burns VVV as a separate line item, on top of the monthly buyback. Against emissions cut to 3M VVV a year (from 14M at launch) and roughly 42% of effective supply already sitting at the burn address, the deflationary backdrop holds. Whether the new credit-purchase burn fully offsets the VVV that DIEM burns release is the open question, and it’s the one to watch.

One caveat carries through all of it. Venice sets the target, the pricing and the emission schedule off-chain. None of those parameters sit behind on-chain governance or a timelock, so the self-correcting curve is only as reliable as the company steering it.

Biggest movers this week

7-day change as of 2026-08-01. The tape skewed red: outside NuNet’s thin 20% pop, gains stayed under 8%, while three names shed more than a fifth of their value.

Gainers:

• NuNet (NTX) +20.8%

• Sentient (SENT) +7.5%

• Oasis Network (ROSE) +5.1%

• Auki (AUKI) +2.1%

• Bittensor (TAO) +1.0%

Fallers:

• Giza (GIZA) -43.6%

• Heurist (HEU) -37.2%

• Vana (VANA) -21.9%

• OpenServ (SERV) -21.6%

• Allora Network (ALLO) -19.7%

See all movers: ownyourmind.ai/projects/movers/

On the watchlist

• Whether anyone mints into the new DIEM headroom. The target step only matters if the supply follows. The rise to 40,000 adds about 2,000 DIEM of headroom, and filling it means locking well over 1M more sVVV, so the headroom is capacity, not a forecast.

• The credit-purchase burn offset. The $5-per-$100 stream is the counterweight to sVVV released by DIEM burns. Watching whether it nets deflationary week to week is the real test of Venice’s flywheel.

• peaq’s Economics 2.0 transition. Two economic systems are running in parallel while stake migrates to trust validators. The migration holding cleanly, or not, is the thing to check before the old model switches off.

From the site

• The DIEM deep-dive carries the full mint-curve mechanism behind this week’s centrepiece, with the live on-chain mint-rate history and a lock calculator: ownyourmind.ai/tokenomics/diem-venice-perpetual-ai-credit/.

• The Venice review tracks the three VVV burn streams and the emission-cut schedule on-chain: ownyourmind.ai/projects/venice/.

• This week’s essay puts a number on a heavy Claude habit. A month on Claude Max, A$160, bought about 44 times that in tokens at Anthropic’s rates. But the cheap price hides two costs: the data you hand to a closed model, and renting access you could otherwise own. It sets both against Venice, where privacy is checkable and DIEM lets you hold the access: ownyourmind.ai/journal/what-a-heavy-ai-habit-costs/.

• The peaq review has a podcast episode this week, useful background before Economics 2.0 settles: ownyourmind.ai/projects/peaq/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds NEAR and VVV.

Here’s the decentralised-AI read worth your time this week. The platform layer sells shared control, so we scored the four platform networks we track on one Freedom dimension: governance decentralisation, which asks who actually decides on upgrades, treasury and protocol direction. The scores run from 2 to 12 out of 20, and the gap is the story. Governance architecture exists on paper for most of them, while operational control stays concentrated.

What shipped this week

• NEAR shipped network upgrade 2.13 to mainnet, adding quantum-safe signing and turning on dynamic resharding. Post-quantum keys use the NIST-approved FIPS-204 scheme, and users switch with a single on-chain transaction. Dynamic resharding lets shards split automatically as state grows, removing a manual validator vote that was a standing centralisation vector. NEAR via PR Newswire, 20 July 2026

• Flux shipped the first mobile build of CumulusVPN, its decentralised VPN, tagging version 1.0.0 for iOS with a single runtime and multi-hop parity with the desktop client. It extends the offering from the browser to the phone. Flux (RunOnFlux) release, 21 July 2026

• Grass opened Season 2 airdrop claims and launched a built-in non-custodial wallet. Rewards pay out in USDC rather than fresh tokens, so the claim adds no new supply, and the wallet drops the seed-phrase step using passkeys or email sign-in. Some long-term contributors are unhappy that the rewards come as USDC instead of more GRASS. Grass via CoinMarketCap, 23 July 2026

This week’s scorecard: who governs the platform layer

We scored the four platform networks we track on governance decentralisation, the Freedom dimension that asks who decides on upgrades, treasury and protocol direction. A vote counts for little if a handful of insiders hold most of the tokens, so the dimension weighs how concentrated the votes are, beyond whether a vote exists at all. Scores are out of 20, grouped most to least decentralised. The band rates this single dimension on its own, separate from each project’s overall Freedom grade.

Working mechanisms, heavy influence (11 to 12):

• Morpheus, 12 : no company or foundation, with an MRC proposal system and Snapshot voting for MOR holders. Held back by a Cyfrin audit showing contract owners retain power over deposited tokens, pseudonymous founders, and one lead technologist’s heavy sway.

• NEAR, 11 : House of Stake and veNEAR give stake-weighted, time-locked voting, and the NEP process gives upgrades a structured path. But the October 2025 inflation halving went through after the community vote missed its required threshold, and the Foundation still drives strategic direction.

Structured on paper, concentrated in practice (9):

• Bittensor, 9 : a bicameral Triumvirate and Senate, and dTAO shifted emissions from a validator oligarchy toward a market mechanism in early 2025. The Opentensor Foundation still controls block production, once used root stake to crash a subnet, and the top 1% of wallets hold roughly 90% of stake.

No mechanism at all (2):

• Intelligent Internet, 2 : every decision rests with a single company director. There is no token voting, no DAO, and no community process. The whitepaper promises progressive decentralisation, but none of it has shipped yet.

The ten-point spread from Morpheus to Intelligent Internet is the familiar decentralised-AI pattern: the more mature projects have functioning mechanisms, and even those carry a founder or foundation with a long reach. This week gave a live example of how control concentrates. On 21 July an exchange turned on custodial TAO staking, routing user stake through a validator run by Yuma, Digital Currency Group’s Bittensor subsidiary. More stake flowing to one operator, on a chain where the top 1% already hold most of it, pulls the wrong way for governance whatever it does for convenience.

Biggest movers this week

7-day change as of 2026-07-26. Risk came off across the board this week: the top gainer added under 9%, five names fell more than 14%, and Ora handed back most of the doubling it posted last week.

Gainers:

• peaq (PEAQ) +8.9%

• Heurist (HEU) +7.8%

• Openmind (ROBO) +6.2%

• OriginTrail (TRAC) +4.9%

• Gensyn (AI) +3.7%

Fallers:

• Ora Protocol (ORA) -29.2%

• OpenServ (SERV) -27.0%

• Allora Network (ALLO) -23.0%

• Giza (GIZA) -16.9%

• Auki (AUKI) -14.5%

See all movers: ownyourmind.ai/projects/movers/

On the watchlist

• Bittensor’s stake concentration. Whether exchange-routed custodial staking deepens the concentration on a chain where the top 1% of wallets already hold most of the stake. It’s checkable on-chain, and it bears on both the governance score and access.

• Intelligent Internet’s decentralisation roadmap. The whitepaper promises a path from single-director control to a multi-sig and an oracle council. Nothing has shipped. If any of it does, it’s the one thing that moves a governance score sitting at 2 out of 20.

• Post-quantum adoption on NEAR. The upgrade is live, but the value is in the take-up. Whether wallets and apps actually migrate users to the new keys is the real test of the release.

From the site

• Akash, rechecked on-chain. We reconciled Akash’s burn-mint-equilibrium against the chain: the AKT leaving circulation sits in the burn vault as backing rather than being destroyed, so the on-chain permanent-burn counter reads zero. The review and our integrity ledger now carry that as a dated match: ownyourmind.ai/projects/akash/.

• The companion explainer, what burn-mint-equilibrium changes for AKT, has a fresh podcast episode this week.

• The Agent Trust Index indexed every ERC-8004 agent we could find across Ethereum and Base and found about half a percent clear all four evidence checks. It reports observations with scope, not scores: ownyourmind.ai/agent-trust/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds GRASS, MOR and NEAR.

A usage record you can actually verify

Here’s the decentralised-AI story worth your time this week. Morpheus inference throughput hit an all-time daily record in mid-July, and unlike almost every “record usage” claim in this sector, the numbers settle on-chain, so we checked them ourselves rather than trusting a dashboard. The demand for owning your inference is real, and it is growing.

What shipped this week

• Aethir shipped agent-to-agent payments on Aethir Claw. Hosted agents can now commission and pay each other, with settlement recorded on-chain before the work is delivered, per Aethir’s own announcement on 15 July. It is an early piece of the agents-paying-agents economy, and worth watching as a mechanism rather than a metric. Aethir blog, 15 July 2026

• Olas launched Toolsmith, a self-improving agent that automates the data analysis, root-cause checks and code fixes that keep prediction agents running on Olas Predict. Steps that needed a human before now run on their own. Olas, 15 July 2026

• Render added a spending surface. RENDER can now be used inside OTOY Studio for AI image and video creation tools, extending the token beyond render-job settlement. Render via CoinMarketCap, 14 July 2026

• Bittensor saw its first notable subnet handover of the summer. Quasar (SILX AI) acquired Subnet 27 and relaunched it as Orion, a training-data subnet that will also feed other subnets. A subnet changing hands is itself a sign the subnet market is becoming liquid. tao.media, 14 July 2026

The spotlight: Morpheus, checked on-chain

Morpheus is a marketplace where you rent inference from independent providers, open-weight and closed frontier models alike, and every session settles on-chain. That last part is why it leads this week. When the team said usage had just set a record, we could test it against the chain instead of taking their word for it.

Our own index of on-chain sessions shows attested throughput climbing every month since February and reaching its highest single day in mid-July. Open-weight coding traffic drove it, with GLM 5.2 the most-used model on the network. The router now carries more than 150 models, including closed frontier models and image and video generators for the first time. On 15 July the node shipped a release that hardens the privacy guarantee at the routing layer: a provider’s bid stays unhealthy until its hardware attestation passes.

Now for the nuance. Throughput and revenue are different things here. Compute on Morpheus is still close to free, which is great for users and no doubt part of the demand story. But the protocol’s only real income is captured staking yield, which has shrunk with deposits.

The supply side is thin, which is natural at this stage: a handful of active providers, one serving most of a day’s tokens, and much of the demand routing through one or two gateways. So the demand for owning your inference is showing up on-chain and growing. The real test is whether that pulls more supply onto the protocol.

Full read, refreshed this week: ownyourmind.ai/projects/morpheus/.

Biggest movers this week

7-day change as of 2026-07-19. Ora roughly doubled on the week and Morpheus added about 4% as its usage record landed, while Openmind led the falls.

Gainers:

• Ora Protocol (ORA) +101.3%

• Giza (GIZA) +28.8%

• Allora Network (ALLO) +28.5%

• OriginTrail (TRAC) +16.0%

• Morpheus (MOR) +4.0%

Fallers:

• Openmind (ROBO) -15.8%

• NuNet (NTX) -14.8%

• Grass (GRASS) -11.4%

• Sahara AI (SAHARA) -10.7%

• io.net (IO) -10.4%

See all movers: ownyourmind.ai/projects/movers/

On the watchlist

• Morpheus’s inference demand. Whether Morpheus makes another all-time high month of inference usage, and whether the active-provider count broadens as demand pulls new supply in. Both are checkable on-chain, and both are strong growth signals.

• Gensyn’s Korean listing. Upbit listed Gensyn (AI) on 30 June with won pairs, opening direct access for Korean retail. It isn’t in our Returns evidence yet, and we’re checking whether it lifts liquidity enough to matter.

• Nous Research’s raise. Multiple outlets report Nous is in talks to raise $75 million or more at a $1.5 billion valuation, on top of roughly $65 million already raised. The figure is report-stage, not confirmed by Nous. If it closes, it reprices the open-weight agent stack around Hermes.

From the site

• Our Agent Trust Index is live. We indexed 95,431 ERC-8004 agents across Ethereum and Base and checked each against four evidence signals: live, paid, sustained and broadly used. Only 526, about half a percent, clear all four. It reports observations with scope, not verdicts: ownyourmind.ai/agent-trust/.

• The companion guide, Four Checks Before You Trust an Agent, walks through verifying an agent before you let it act on your behalf.

• The Morpheus review, refreshed this week with the on-chain demand read: ownyourmind.ai/projects/morpheus/.

• The OYM podcast is on Spotify, Apple Podcasts and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds GRASS, MOR and RENDER.

Five things shipped across the DeAI cohort this week, and two were memory infrastructure for AI agents. OriginTrail rebuilt its knowledge graph as a three-layer memory system, and Vana shipped a personal memory server you run yourself. The releases kept coming even as prices sold off, with the week’s biggest moves all to the downside. Here’s what landed, and what didn’t.

What shipped this week

• OriginTrail rebuilt the graph for agents. DKG V10 went live on mainnet on 25 June 2026 across Base, Gnosis and NeuroWeb, and has shipped steadily since, with v10.0.5 landing on 9 July. It reframes the Decentralised Knowledge Graph as three-layer memory: a free local working layer, a free shared layer gossiped between agents, and a paid on-chain layer where knowledge becomes cryptographically verifiable. Only that last layer costs TRAC, so the token now sits at the point where memory becomes provable rather than at every write. OriginTrail DKG releases · our review

• Vana shipped portable AI memory. On 8 July 2026 it launched a public beta of the Vana app plus a personal data server that runs on your own device and exposes an MCP endpoint, so your memory travels with you across AI platforms instead of living inside one. It also folded in a memory-protocol team by acquisition. No tokenomics change, but it sharpens the data-sovereignty thesis. Vana blog, 8 July 2026 · our review

• Render wired RENDER into SaladCloud. Salad integration milestones 1 and 2 went live on 9 July 2026: customers can fund SaladCloud compute with RENDER, and GPU providers can redeem earnings in RENDER to Solana wallets. The next phase folds in the full burn-and-mint step, which is where the token impact lands. Render, 9 July 2026 · our review

• Grass passed a revenue-sharing vote. A 7 July 2026 governance vote approved routing a portion of Grass’s reported ~$33M annual USDC revenue to GRASS holders, the first structural link between the network’s enterprise revenue and the token. The market treated it as sell-the-news: GRASS was among the week’s biggest fallers (see movers below). CryptoBriefing, 3 July 2026 · our review

• Underneath the headlines, the cohort kept shipping code. Release feeds show Morpheus’s node at v7.5, NEAR’s nearcore at 2.13, a Walrus mainnet update, and steady releases from Autonolas, Ocean, IoTeX and Flux. Version bumps rather than headline events, but the build cadence held up while prices fell.

One thing didn’t ship: Grass’s Season 2 native wallet. It was announced for the first half of 2026, the half closed on 30 June, and the wallet still isn’t live. Incentives keep vesting on the emission schedule while user distribution waits on the wallet.

Biggest movers this week

The week ran risk-off. The five largest moves were all falls, led by Warden Protocol shedding nearly half its value, while the top gainer managed just over 8%.

Gainers:

• Heurist (HEU) +8.2%

• Ora Protocol (ORA) +6.2%

• Morpheus (MOR) +5.6%

• Allora Network (ALLO) +4.9%

• Olas (OLAS) +1.8%

Fallers:

• Warden Protocol (WARD) -49.0%

• OpenServ (SERV) -28.5%

• Grass (GRASS) -22.4%

• peaq (PEAQ) -19.1%

• Auki (AUKI) -18.9%

7-day change as of 13 July 2026. See all movers: ownyourmind.ai/projects/movers

On the watchlist

• OriginTrail’s audit gap. V10 shipped new smart contracts, still without a published audit from a major firm. That gap grows as staking value sits on the new code; it’s the number we’ll re-read next month.

• Grass revenue sharing, from vote to flow. The vote approved the mechanism, but no stablecoin yield has reached holders yet. Whether it actually flows is the test of the Grass value-accrual story, and the reason the token sold the news.

• Vana’s beta retention. A personal memory server is a strong sovereignty story on paper. The signal worth watching is whether the public beta keeps users once the novelty fades.

From the site

• Our latest journal piece: The State of Agent Trust, on how to judge whether an autonomous agent can be trusted with a wallet and a task.

• The OYM podcast is on Spotify, Apple Podcasts, and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds GRASS, MOR, NEAR, and RENDER.

Here’s the number worth your attention in decentralised AI this week. In 30 days, AI agents settled 12 million payments on Base, five times the month before, while the total dollars they moved actually fell. The agent economy is active and growing, but the value is fragmenting into fractions of a cent. The week’s other headline sharpens the same question from the equity side: Venice became a $1 billion unicorn on a raise its token holders have no claim on.

Subscribe now

What shipped this week

• Venice became a unicorn. It closed a $65 million Series A at a $1 billion valuation on 1 July, led by Dragonfly with Coinbase Ventures and North Island Ventures, its first external raise after self-funding since the January 2025 airdrop. The catch for token holders: the round is equity in the company, not a VVV sale, so it adds no token dilution but introduces an equity preference stack ahead of VVV holders for the first time. On the same day, the scheduled emission cut to 3 million VVV a year took effect, the latest in a run of reductions from 14 million in fourteen months. TechCrunch, 1 July 2026

• NEAR opened a euro on-ramp on near.com, letting EU users move euros on-chain via SEPA Instant and trade confidentially across crypto markets from their own wallets. It extends the confidential-finance surface NEAR has been building rather than breaking new ground, but it widens the addressable base for the Intents routing layer underneath. near.com via PR Newswire, 1 July 2026

This month’s x402 reading: more settlements, less money

x402 is the rail that lets an AI agent pay for compute, data, or inference with no human in the loop. It repurposes the HTTP 402 “Payment Required” code: the server quotes a price, the agent pays in USDC on Base, no account and no API key. We track settlement on it from on-chain data. The rolling 30-day record just shifted in a way worth reading.

Comparing the 30 days to 4 July against the prior 30:

• Settlement count rose 405%, from 2.4 million to 12.3 million EIP-3009 transfers across the facilitator pools we track.

• Dollar volume fell 21%, from $7.1 million to $5.6 million in USDC settled.

• Average settlement size collapsed from $2.94 to $0.46, an 84% decline.

• Unique payers dropped 36%, but each transacting wallet settled far more often, from roughly 14 to 108 times a day.

• Unique payee addresses spiked on 2026-06-29, from 831 the day before to 2,346, and stayed elevated through 2026-07-03.

The structural read is a shift to production micropayments. When settlement count moves fivefold while the dollars fall, agents are paying in smaller, more frequent increments, which is the signature of software transacting on a schedule rather than humans making purchases. The payee spike is the signal to watch: a jump in the number of addresses being paid points to new service endpoints coming online, and service-side diversity is a leading indicator of how deep the agent economy actually runs.

One caution keeps this honest. A higher settlement count at lower value does not equal more economic activity, and here the dollar volume actually fell. The count surge could be a single large user switching from batch settlement to per-call payment, inflating the number without adding demand. x402’s settlement data doesn’t expose service-level detail, so the count is a leading indicator, not proof the agent economy has matured.

Worth re-reading in the August window. Full mechanism breakdown: When AI Agents Hold Wallets.

Biggest movers this week

7-day change as of 2026-07-05. A split week: training and agent tokens led the risers, while the fallers were projects with recent structural problems. Giza led the falls, down 37%, extending the slide that followed our June re-score when its flagship agents were retired. Venice fell 13% in the same week it raised $65 million, a reminder that the token isn’t the equity.

Gainers:

• OpenServ (SERV) +32.5%

• Gensyn (AI) +31.9%

• Allora Network (ALLO) +19.8%

• Auki (AUKI) +12.9%

• Sentient (SENT) +12.8%

Fallers:

• Giza (GIZA) -37.3%

• Ora Protocol (ORA) -23.1%

• Phala Network (PHA) -16.1%

• Openmind (ROBO) -14.5%

• Venice (VVV) -12.9%

See all movers: ownyourmind.ai/projects/movers/

On the watchlist

• The x402 payee spike. The jump to 2,346 unique payee addresses on 29 June held for five days. If it sustains into August, it points to real service-side growth on the agent-payment rail. If it fades, it was a one-off. This is the number we will re-read next month.

• Venice’s value-accrual question. The Series A introduces an equity claim ahead of the token for the first time. How much that should weigh on Venice’s value-accrual score is under review this month. The raise is good for the company; its effect on VVV holders is the open question.

• Allora’s Forge launch. Allora announced Forge, an arena where predictive models compete and earn, on 2 July, and the token rose about 20% on the week. The launch is so far carried only on the project’s own channels and a PR wire, so we are holding it for independent corroboration before scoring anything.

From the site

• The companion to this issue: When AI Agents Hold Wallets, on why agents holding wallets changes the payment layer, and why x402 captures none of the value it moves.

• The full Venice review, updated for the Series A: ownyourmind.ai/projects/venice/, including the token-versus-equity distinction the raise sharpens.

• The OYM podcast is on Spotify, Apple Podcasts, and the major directories under “Own Your Mind”. The latest episode is on Phala.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai ·

Disclosure policy

Disclosure: OYM holds NEAR and VVV.

Decentralised compute sells a simple promise: no company owns the machines. The hardware lives up to that more often than the control does. This week we scored the 14 compute networks we track on one Freedom dimension, infrastructure decentralisation, which asks who runs the nodes, who owns the hardware, and whether anyone can join without permission. The scores run from 6 to 16 out of 20, and the gap is the story.

Subscribe now

What shipped this week

• io.net turned on Stripe and x402 payment support for Agent Cloud, so AI agents can provision and pay for GPU compute with no human in the loop. It expands the product while the core orchestration platform stays centralised. Our Crypto Talk, 24 June 2026

• Phala was named a launch partner for the open-source GLM-5.2, running the model in a TEE inference stack on Intel TDX and NVIDIA H100 enclaves. It strengthens the confidential-compute offering without changing Phala’s early-stage infrastructure posture. Phala, 16 June 2026

• Sahara AI deferred most of its 26 June token release after a 60% price crash and insider-sale accusations the team denies. Founder, team and advisor tokens move back six months and investor tokens three, a voluntary vesting extension covering the bulk of a 1.03 billion SAHARA tranche, about 10% of supply. CryptoRank, 24 June 2026

This week’s scorecard: who runs the compute

We scored the 14 compute networks we track on Infrastructure decentralisation, the Freedom dimension that asks who runs the nodes, who controls the hardware, and whether anyone can participate. A high score means no single entity has to stay up for the network to work. Scores are out of 20, grouped most decentralised to least. The band rates this single dimension, not each project’s overall Freedom grade.

Broadly distributed (14 to 16):

• Flux, 16 : roughly 7,000 nodes across 67 countries, permissionless node operation.

• Morpheus, 15 : permissionless compute marketplace, independent providers, multi-chain.

• Nosana, 14 : permissionless GPU hosting, large registered base.

• Golem, 14 : peer-to-peer network, anyone joins as provider or requestor.

Permissionless, with central vectors (9 to 12):

• Akash, 12 : permissionless deployment, validators and providers tracked live.

• Theta, 12 : BFT consensus with enterprise validators and roughly 10,000 edge nodes.

• Flock, 11 : permissionless nodes and 262 validators for multi-party training.

• Gensyn, 10 : permissionless testnet for decentralised training.

• io.net, 10 : GPUs sourced from 138-plus countries, centralised core platform.

• NuNet, 10 : permissionless device onboarding over an open-source libp2p mesh.

• Aethir, 9 : large container footprint, closed-source orchestration layer.

• Heurist, 9 : open-source miner software, permissionless registration.

Permissioned or early-stage (6 to 8):

• Phala, 8 : TEE inference on Intel TDX, still an early-stage L2.

• Render, 6 : roughly 15,670 registered operators, but the network is permissioned.

The hardware is spread out. The control stays concentrated.

The top and bottom of the table separate on one question: can anyone run the infrastructure without asking. Flux, Morpheus, Nosana, and Golem sit at 14 to 16 because node operation is open and the hardware base is distributed. Flux at 16 is the highest infrastructure score we record across the entire project set, compute or otherwise. Render and Phala sit at 6 to 8 because their networks are either explicitly permissioned or still early-stage, with limited operator autonomy.

The middle band is where the marketing and the architecture part ways. io.net and Aethir both report large geographic footprints, 138-plus and 94 countries respectively, and both land at 9 to 10. The hardware is distributed; the orchestration that decides what runs where is closed-source and operated by a single entity. A wide map isn’t the same as decentralised control, and the scorecard is built to tell them apart.

The other thing the table shows is how little moves. None of these infrastructure scores has shifted in our recent reviews. io.net’s overall Freedom score rose a point in mid-June, but on transparency, the epoch explorer now publishes burn data, not on infrastructure. A busy month of product launches and partnerships changed their capabilities. Control stayed where it was.

Biggest movers this week

7-day change as of 2026-06-28. A red week across the board: only three of the 39 DeAI tokens we track closed up. Ora led the gainers; agent tokens led the falls. The data layer was the exception, Grass and Ocean making up two of the only three tokens in the green.

Gainers:

• Ora Protocol (ORA) +33.0%

• Grass (GRASS) +23.0%

• Ocean Protocol (OCEAN) +12.5%

Fallers:

• OpenServ (SERV) -31.2%

• Allora Network (ALLO) -23.2%

• Warden Protocol (WARD) -20.1%

• OriginTrail (TRAC) -19.8%

• Theta Network (THETA) -18.0%

See all movers: ownyourmind.ai/projects/movers/

On the watchlist

• Render’s permissioned bottleneck. Render sits at the bottom of the scorecard because the network decides who joins. Any move to permissionless node operation would be the single change that lifts the 6 out of 20. Nothing this cycle.

• io.net’s closed core. Agent-paid GPU compute through x402 is a real product step, but the orchestration layer stays closed-source. Autonomous demand actually settling on the rail is the thing to watch, not the integration itself.

• Phala’s early-stage exit. The GLM-5.2 partnership strengthens the confidential-compute story. The infrastructure score only moves when Phala progresses past its current L2 stage.

From the site

• The companion to this scorecard: How compute networks accrue value. Infrastructure decentralisation is one axis; whether the token captures any of the value is the other.

• Who actually earns running this hardware: the GPU provider income guide, across cash and token networks.

• The OYM podcast is on Spotify, Apple Podcasts, and the major directories under “Own Your Mind”. The latest episode is How to Earn with RENDER.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds AKT, GLM, GRASS, MOR and RENDER.

io.net shipped the headline tokenomics change it had promised for months, and the launch came wrapped in a confident revenue story. The burn is now live and settling on-chain, which is real progress on transparency. The funding behind the burn is where the campaign and the ledger part ways. A live burn and a revenue-funded burn aren’t the same thing, and the difference is the whole investment case.

What shipped this week

• io.net activated the Incentive Dynamic Engine on-chain, with the burn component live and settling on Solana since 1 June 2026 and a launch campaign on 11 June claiming an $8M enterprise deal and 4 billion AI tokens processed a day. io.net, the IDE is now live, 11 June 2026

• IoTeX activated its “Yap” hard fork (v2.4.0/2.4.1) on 7 June 2026, its largest mainnet upgrade of the year, bringing Ethereum Pectra compatibility and native account abstraction. A platform upgrade, not a tokenomics or decentralisation shift. IoTeX upgrade, 7 June 2026

The burn is live. The revenue funding it isn’t.

Give io.net the credit it earned. The burn is deployed, it settles on-chain, and the epoch explorer now publishes per-epoch emission and burn data anyone can check. That’s a decent step up from a roadmap promise, and it’s why we nudged io.net’s Freedom Score up a point, to 39 out of 100, when we re-scored the project on 17 June. The transparency gap narrowed.

The funding claim is the problem. io.net frames the burn as powered by network revenue, scarcity earned from real usage. The explorer tells a plainer story. Per-epoch emissions are flat, so the demand-driven half of the design isn’t live yet, and on-chain client purchases run at roughly 3% of what suppliers are paid. A burn drawn from that gap is funded by emissions, not by revenue. The mechanism is honest about destroying tokens. It isn’t yet honest about where the money comes from.

The headline numbers lean the same way. The $8M deal and the 4-billion-tokens-a-day figure are io.net’s own reporting, repeated through a paid press release, not independently verified. Independent trackers put annualised revenue well below io.net’s own number, which is why we marked Revenue Sustainability down even as the supply mechanics improved. The two moves cancelled out: Returns held flat at 54 out of 100, a D.

What it actually is

Strip the campaign and io.net is a centralised GPU marketplace with a token that is more transparent than it was on 1 June. That’s better than a token with no live burn. It isn’t a decentralised moat. The core platform stays closed, governance stays absent, and the project is still squeezed between Akash on the open side and the hyperscalers on the centralised side. The IDE helps the token story without touching that position.

The bull case is now narrower and cleaner to test. It needs the burn ledger to keep tracking real usage on the explorer, the flat emissions to flip into the demand-driven schedule that was promised, and the enterprise pipeline to turn launch-day language into signed contracts. All three are checkable in public over the next few weeks. That’s the upside of shipping on-chain: the claims stop being a matter of trust.

Biggest movers this week

7-day change as of 2026-06-21. io.net itself was among the week’s gainers even as our read on its burn stayed cautious, a reminder that the market and the scorecard often disagree. Olas led the board; Ora and OpenServ fell hardest.

Gainers:

• Olas (OLAS) +30.7%

• io.net (IO) +13.3%

• Warden Protocol (WARD) +11.6%

• Heurist (HEU) +10.6%

• Grass (GRASS) +8.4%

Fallers:

• Ora Protocol (ORA) -28.9%

• OpenServ (SERV) -27.7%

• Nosana (NOS) -15.7%

• Bittensor (TAO) -13.9%

• Sahara AI (SAHARA) -13.9%

See all movers: ownyourmind.ai/projects/movers/

On the watchlist

• io.net’s burn ledger. The number to watch is whether burn volume tracks actual usage on the explorer, or stays flat while emissions fund it. The first few IDE epochs will tell.

• io.net’s second enterprise deal. A signed, public contract would move the revenue read. “Advanced stages” language doesn’t.

• The demand-driven emission switch. io.net designed emissions to respond to demand. Until per-epoch issuance stops being flat, that half of the IDE is marketing, not mechanism.

From the site

• The full io.net scorecard, re-scored after the burn went live: ownyourmind.ai/projects/io-net.

• Where io.net sits against its nearest rivals: Render vs Akash vs io.net.

• The OYM podcast is on Spotify, Apple Podcasts, and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds AKT and GRASS.

Subscribe now

Three things shipped this week, and together they map three layers of decentralised AI maturing at once. Aethir, a GPU network, launched its own LLM API and now sells the inference instead of only renting the hardware that runs it. Akash shipped a base-layer upgrade that is live and burning. Allora put decentralised AI inference behind live trades on a regulated US prediction market. Same sector, three different bets on where the value sits.

What shipped this week

• Aethir launched Aethir Mesh on 3 June 2026, a proprietary API layer serving open-weight models (DeepSeek V4, Kimi K2.6, GLM-5.1, Qwen3.6 and others) on its own decentralised GPU network. Managed access to open weights, closed operator. Aethir Mesh, launched 3 June 2026

• Akash Network activated its Mainnet 18 upgrade (v2.1.0) on 11 June 2026, adding Oracle v2 and resource reclamation (AEP-82) on top of the burn-mint loop that went live in March. Upgrade notes, 11 June 2026

• Allora Network began executing live trades on Kalshi, a CFTC-regulated prediction market, on 1 June 2026, through the Cobot product built on its inference network. PredictionNews, 1 June 2026

Aethir climbs the stack

The sharpest move of the week is Aethir selling inference. A GPU rental network that packages its own LLM API is competing with the customers it used to rent GPUs to. For users it’s convenient: managed access to a shelf of open-weight models without provisioning a single node.

The catch is sovereignty, the whole reason most people look at decentralised AI in the first place. Aethir Mesh is a hosted, proprietary API on a centralised operator, and we rate Aethir 3 out of 10 on our Freedom Score. It sits at the convenience end of the spectrum, where an OpenAI endpoint sits, and it does nothing for the control problem that self-hosting solves. Useful if you want open weights without the ops, and little help if you wanted to own the stack.

The same category, the opposite design

Akash is in the same compute bracket and made the opposite bet. Its Mainnet 18 upgrade hardens an Apache-2.0 base layer (we grade Akash a C, 66 out of 100 on Freedom) where the burn-mint loop now routes compute spend through an on-chain AKT buy. The scale gap is still the honest story here. At current lease revenue, burns offset only a fraction of issuance, so this is upgrade cadence rather than a step change. The point is the direction: Akash hardens the open layer, Aethir packages a closed product on top.

Allora is the application-layer signal. Decentralised AI inference output reaching a regulated venue is the kind of real-world endpoint the sector keeps promising and rarely ships. Worth one caveat on the token: a 48.55% insider cliff arrives in November 2026, which on our read could nearly double the circulating float in a single event, against revenue we still can’t verify.

Three layers, three sovereignty profiles. The week’s signal is that compute providers are no longer content to rent hardware. Some are climbing toward the product, and which way they climb is the thing to watch.

On the watchlist

• Whether Aethir Mesh opens. The change that would move its Freedom Score is model portability or a self-host export path. A proprietary endpoint on a centralised operator is the convenient option, short of sovereign.

• Akash burn scale. Oracle v2 and resource reclamation harden the base, but the number that matters is whether burn volume ever catches issuance. It hasn’t yet.

• Allora’s November cliff. A 48.55% insider cliff, roughly five months out, lands against revenue we can’t verify. That’s the risk to size before the technical story.

From the site

• New review: OpenServ (SERV). Credible research and a properly open SDK wrapped around a closed reasoning engine and a buyback-burn that has never run on-chain. Freedom 24, Returns 34, both an F.

• The OYM podcast is on Spotify, Apple Podcasts, and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds AKT.

Subscribe now

Two pieces of plumbing shipped this week: Virtuals swapped its cross-chain settlement layer, and io.net opened a Korean fiat pair. Useful, neither one a surprise. The two mechanisms worth your time both sit in the tokenomics layer: Morpheus builder subnets, a funding mechanism your staking choice can pull in different directions, and x402, the agent-payment rail that proves agents transact at scale while capturing none of it. Here is how each one works.

What shipped this week

• Virtuals Protocol migrated VIRTUAL’s cross-chain messaging from LayerZero to Chainlink CCIP, across Base, Solana, Ronin, Arbitrum, and the XRP Ledger. The settlement layer changed; the inference layer did not. PRNewswire, 4 June 2026

• io.net listed IO on Upbit’s KRW market, adding a direct Korean fiat on-ramp alongside its existing Binance and Coinbase pairs. Wider exchange access, same closed-source core. BloomingBit, 29 May 2026

What your stake actually does

Morpheus builder subnets are a funding mechanism, and the closest thing to them in design is Gitcoin Grants. You stake MOR toward a subnet, the protocol directs a slice of the capped, scheduled emission to that project, and the builders contract pays it to the operator’s chosen wallet. MRC 22, the proposal that set out subnet staking, has stakers earn the backed project’s token, says staking is at your own risk, and has Morpheus review and endorse nothing.

The open design means there is no wrong way to stake. There are three reasons people do, and they land differently on MOR:

• For yield. Some operators pass MOR back as a headline rate. This is the familiar choice and the least remarkable, because it’s yield farming. Sell the rewards and you pipe the subsidy into sell pressure; restake them and you at least lock supply back up.

• For access. Stake to use a product. The Marketplace API gateway gives you inference access, no yield in the loop, and it sits among the most-staked subnets (June 2026). Real demand, locked supply, no sell pressure, the healthiest pattern for the token.

• For a public good. Back a project that pays you nothing directly because it strengthens the network, which lifts MOR for every holder indirectly. This is the Gitcoin-donor case the design was written for.

The trouble is the dashboard lists a harvester that ships nothing beside the gateway with the same furniture, and weights stake linearly so one whale counts like a hundred backers. Gitcoin solved exactly this, with qualification review, breadth-weighted matching, and self-dealing detection. Morpheus shipped without the guardrails.

This matters more as agents arrive. An allocator agent optimises the only number it can read, the posted yield, and routes to whoever games it best. That is the race to the bottom: point enough agents at an unguarded dashboard and capital floods the harvesters, while the access and public-good stakes that actually compound MOR look least like a yield product.

Nobody is staking wrong. The choice just decides what the capped emission ends up building.

Full analysis: Morpheus Builder Subnets Are Gitcoin Grants for Decentralised AI.

The other mechanism: x402 and the value-capture gap

Coinbase’s x402 protocol lets an AI agent pay for an API call in stablecoins over plain HTTP, with no account and no API key. It’s the cleanest read we have on whether agents are actually transacting, and the answer is yes. Our on-chain tracker of Base relayer settlements shows monthly throughput has more than doubled since the protocol’s first full month, with the active relayer set growing from a handful at launch to dozens.

Settlement counts are climbing faster than dollar volume, so ticket sizes are shrinking as more services plug in. Figures refreshed 2026-06-07.

The catch is value capture. x402 runs on Coinbase-subsidised gas and settles in USDC, so the activity accrues to no token directly. Artemis showed earlier this year that self-reported x402 figures overstated the on-chain reality by roughly 15 times, and that gap hasn’t fully closed.

Growing payment volume is the agent-commerce thesis maturing well ahead of any moat. The mechanism to watch is whether a fee-bearing version ever appears, because subsidised throughput is a usage signal, not value capture. How agent wallets actually work.

On the watchlist

• A labelling layer for Morpheus subnets. The one change that fixes the mental-model problem is a marker separating subnets that fund public goods from subnets that extract. Nobody has shipped it.

• Virtuals’ inference path. The CCIP migration hardens cross-chain settlement, but GAME inference still runs on Virtuals-hosted servers. A permissionless inference option is the change that would move the Infrastructure score.

• A fee-bearing x402. The single change that would turn x402’s throughput into token value is a version that charges a settlement fee instead of running on subsidised gas. Coinbase hasn’t signalled one.

From the site

• New map: which AI agents actually help, and which are just a token. No agent-category project we rate earns better than a C on either axis.

• The OYM podcast is on Spotify, Apple Podcasts, and the major directories under “Own Your Mind”.

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds MOR and VIRTUAL.

The AI agent sector sells autonomy, but the infrastructure underneath splits sharply. Some platforms run agents client-side, with no central server to fail. Others route every decision through their own hosted inference. This week we scored the six agent projects we track on one dimension: who actually runs the infrastructure.

Subscribe now

What shipped this week

• Virtuals Protocol, EconomyOS launched, giving onchain AI agents managed email inboxes, non-custodial wallets, and virtual Visa cards for autonomous commerce. Crypto Briefing, 16 May 2026

• NEAR, Universal Send brought confidential cross-chain payments to near.com: pay anyone in any asset across supported chains without exposing wallet activity. Invezz, 29 May 2026

• NEAR AI, the Government of Bermuda signed on for a prototype assistant that processes citizen data inside hardware-enforced enclaves, the first named NEAR AI customer. The Royal Gazette, 14 May 2026

This week’s scorecard: who runs the agents

We scored the six agent-sector projects on Infrastructure decentralisation, the Freedom dimension that asks who runs the nodes, who controls the hardware, and whether anyone can participate. A high score means no single point of failure. A low score means the token is bolted onto centralised infrastructure. Scores are out of 20, ranked most decentralised to least.

• AntSeed, 14/20 (broadly distributed): client-side execution, no central inference dependency.

• Olas, 12/20 (permissionless, some vectors): permissionless nodes, but a Gnosis Safe multisig controls key contracts.

• Warden, 10/20 (permissionless, some vectors): distributed validators, centrally operated key management.

• Fetch.ai / ASI, 9/20 (permissionless, some vectors): permissionless agents, foundation-controlled flagship services.

• ElizaOS, 6/20 (foundation-controlled): open-source framework, production agents lean on centralised APIs.

• Virtuals, 5/20 (foundation-controlled): GAME inference runs on Virtuals-hosted servers.

The band in brackets rates this single dimension, not each project’s overall Freedom grade.

The spread runs nearly the full range, from AntSeed at 14 to Virtuals at 5, and it tracks one architectural choice. AntSeed and Olas run agents client-side or across permissionless networks, so no central server has to stay up for an agent to work. AntSeed’s 14 is the highest Infrastructure score among the six agent projects we track, though compute-DePIN networks outside the sector score higher still (Flux at 16, Morpheus at 15).

At the other end, ElizaOS and Virtuals score 5 to 6 because their core value, the inference and decision-making, depends on team-operated infrastructure. ElizaOS ships an open-source framework, but production agents still call whatever API the founder’s keys point at. Virtuals runs GAME Cloud as a hosted service. If those endpoints change terms or go down, the agents stop.

This is the gap the EconomyOS launch does not close. Giving agents inboxes, wallets, and Visa cards extends what a Virtuals agent can do, but every one of those actions still routes through Virtuals-hosted inference. More capability, same single point of control.

No Infrastructure score moved in May. No project in the sector shipped a node-permissioning upgrade or an infrastructure migration. The decentralisation posture is static, which is worth saying plainly: a busy week of product launches changed what these agents can do, not who controls them.

On the watchlist

• Virtuals inference path. EconomyOS raises the stakes on GAME Cloud centralisation. A permissionless inference option would be the single change that moves the 5/20.

• ElizaOS production defaults. The framework is open; the deployments are not. Any shift toward self-hosted or attested inference in the default path would lift the 6/20.

• NEAR Universal Send adoption. NEAR reports that close to half of near.com swap volume now routes through the confidential path. An independent audit of that figure would turn a NEAR-reported claim into a verified one.

From the site

• NEAR page refreshed: the Bermuda NEAR AI pilot, Universal Send, and the June upgrade 2.13 (dynamic resharding plus post-quantum signing) are now in the NEAR review, with the scoring rationale for why none of it moves the grade yet

• New tokenomics deep-dive: How Morpheus Builder Subnets Work, how the builder-versus-compute split reshapes the emission flow

• The OYM podcast is on Spotify, Apple Podcasts, and the major directories under “Own Your Mind”

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds NEAR and VIRTUAL.

NEAR is the week's most consequential DeAI project on two fronts at once. Dynamic resharding ships in June 2026, automating shard creation and removing a manual centralisation vector from the sharding architecture. At the same time, Intents fee revenue is closing on the network's annual issuance. The infrastructure story and the value-accrual story converge in the same month.

What shipped this week

• NEAR, dynamic resharding announced for June 2026, automating shard creation without human intervention. CoinDesk, 22 May 2026

• ASI Alliance, Agent Launchpad live on BNB Chain, letting AI agents autonomously deploy tokens and list on DEXs. Crypto Briefing, 20 May 2026

• Cookie DAO, COOKIE added to Binance's monitoring tag list alongside eight other tokens, flagging potential delisting risk. Binance, 22 May 2026

Why NEAR's June matters

Dynamic resharding is the most significant NEAR protocol change since Nightshade 2.0 shipped in May 2025. The current Infrastructure score (14/20) notes that Pagoda runs some core RPC infrastructure and that the Foundation has outsized influence on development priorities. Manual shard management is a centralisation vector: someone decides when and how to add shards. Dynamic resharding removes that human decision, making the infrastructure more resilient and less dependent on Foundation or Pagoda operational choices.

The governance pattern is the catch. The inflation halving in October 2025 was implemented despite the community vote failing to reach threshold. Chorus One called that "a dangerous precedent." If dynamic resharding ships as a Foundation-driven upgrade without a governance vote, the same concern applies: good technical outcome, concerning process.

The Infrastructure dimension may improve from this upgrade. The Governance dimension (11/20) would not benefit from another top-down upgrade path. Both can be true simultaneously, and the score change waits on which path the upgrade actually takes.

The Intents volume growth ($15.2B to $19.17B cumulative in roughly two months) reinforces the Returns story. The fee switch, activated February 2026, converts this volume directly into NEAR demand. The annualised run-rate sits at ~$34.6M per DeFiLlama and is approaching the ~32M NEAR annual issuance at current prices, bringing the protocol close to a net deflationary state. This is one of the most concrete value-accrual mechanisms in the OYM portfolio, and the figure is independently verifiable rather than self-reported.

No score change is warranted yet. Dynamic resharding has not shipped. The Intents trajectory continues the existing Returns momentum already captured in the 2026-04-11 increase (Returns 70 to 73). The question for the next review cycle is whether dynamic resharding ships via governance vote or validator adoption, and whether that distinction moves the Freedom Governance dimension.

On the watchlist

• Dynamic resharding governance path. NEP vote, or Foundation-driven validator upgrade? The answer determines whether the Governance dimension moves alongside Infrastructure.

• NEAR Intents fee run-rate. 30-day fees holding above $2.8M sustains the annualised figure and strengthens the deflationary crossover case. A drop weakens the narrative.

• Cookie DAO 30-day monitoring window. Binance reviews the tag at the end of the period. Delist or reinstate? Either outcome is editorially significant.

From the site

• New tokenomics deep-dive: How Morpheus Builder Subnets Work, how the builder-versus-compute split is reshaping the emission flow

• New journal: Bittensor Conviction Ships, the governance upgrade that landed on 13 May and what it means after the Templar exit

• Recent podcast: Episode 060 (Morpheus Builder Subnets) and Episode 056 (Training without a data centre). The OYM podcast is on Spotify, Apple Podcasts, and the major directories under "Own Your Mind"

• Venice page refreshed for Liquidity & Access (Robinhood, Upbit, eToro, Crypto.com all added), Akash Q1 2026 data refreshed from Messari

Cheers,
Bobski

Own Your Mind. Independent research on decentralised AI. No sponsors.

ownyourmind.ai · @ownyourmindai · @0xbobski

Disclosure policy

Disclosure: OYM holds NEAR.