Security firms have linked a single attacker to two separate exploits that drained roughly $2 million from AI-focused crypto projects Fetch.ai and NuNet, with NuNet’s token collapsing to an all-time low as the stolen funds were laundered into ether within hours of the first drain.
On-chain monitoring firm PeckShield reported that the same wallet moved approximately 8.7 million FET out of Fetch.ai, a haul worth about $1.53 million at the time. The funds came from the project’s token converter, a bridge-like contract on Ethereum that lets users move FET between chains. Blockaid, which also tracked the incident, said the attacker used a permission flaw in the converter rather than a bug in the token itself.
Fetch.ai later traced the breach to a leaked signing key. Roughly 8.7 million FET left the converter in a single signed call, according to the company’s own account of the incident, which security researchers shared in the hours after the drain. The single-call design meant there was no window in which a rate limiter or pausing mechanism could have intervened, and no multisig approval stood between the key and the treasury.
The same wallet then pivoted to NuNet, a decentralized compute project, and minted about 408.5 million unauthorized NTX tokens through the project’s deployer account. Blockaid valued the illicit mint at roughly $452,000 to $463,000 depending on the price feed used. The combined activity across the wallet cluster reached approximately $2.01 million.
NTX paid the heaviest price. The token fell more than 70 percent in 24 hours and touched an all-time low of $0.000328 on September 20 before stabilizing near $0.0004. FET fell about 5 percent over the same period, a smaller move that tracked the broader market sell-off, with total crypto market capitalization down roughly 4 percent on the day. The gap between the two reactions reflects liquidity as much as severity: FET trades deep on major venues, while NTX had thin books that any sell pressure could crater.
The attacker moved quickly to convert the proceeds. PeckShieldAlert reported that the exploiter swapped stolen assets for 546.36 ETH, worth about $1.44 million at the time of reporting, a standard laundering step executed before protocols or exchanges can freeze anything. Tracing firms are now following the ETH through mixer-adjacent addresses, though no exchange freeze has been announced.
SlowMist flagged the missing access control in the converter contract within hours of the drain, pointing to the exact function that released funds without checking limits.
It lacks the checkLimits(amount) modifier present in conversionOut(), and does not verify any on-chain state before releasing funds. – SlowMist, September 20, 2026
The Fetch.ai and NuNet incidents were not isolated. September has been an expensive month for the sector, with cumulative hack losses now exceeding $333 million according to security trackers. Three days before the Fetch.ai drain, the Starknet lending protocol Nostra lost $3.5 million to a manipulated oracle price feed. Earlier in the month, the Liquid sidechain saw 4,000 bitcoin taken in an incident its operators described as a white hat intervention, though the funds remain largely unmoved. MultiversX, another established network, halted its mainnet on September 20 after attackers used an atomicity bug at the virtual machine layer, forcing a fix through a shadow fork.
Two Projects, One Wallet
| Project | Attack vector | Amount | Token impact |
|---|---|---|---|
| Fetch.ai (FET) | Token converter permission flaw, leaked signing key | $1.53 million, 8.7M FET | FET down about 5% |
| NuNet (NTX) | Unauthorized mint via compromised deployer | About $463,000, 408.5M NTX | NTX down over 70%, all-time low |
| Nostra (Starknet) | Oracle price manipulation | $3.5 million | Protocol paused |
| MultiversX (EGLD) | VM-layer atomicity bug | Mainnet halted | Fix in shadow fork testing |
Both Fetch.ai and NuNet sit at the intersection of two of the most heavily marketed sectors in crypto: artificial intelligence and decentralized compute. That positioning attracts capital, exchange listings and media attention, but researchers note it does nothing for contract quality. The recurring pattern across AI-token projects is familiar by now, with ambitious roadmaps, cross-chain infrastructure and permission structures that create single points of failure.
When one contract’s access controls are misconfigured, the blast radius can extend well beyond the original target. NuNet holders discovered this directly, since their project’s deployer was compromised through access that appears to have been obtained in the same operation as the FET drain, even though the two protocols are separate codebases with separate teams. The shared receiving wallet was what let investigators connect the two attacks within a day, and it suggests the attacker studied both projects before moving.
The Audit Question
Security analysts pointed out that the Fetch.ai converter was missing a limit check that its own outbound conversion function carried, an asymmetry that any routine audit review should have caught. The mint path on NuNet likewise relied on a deployer key whose compromise went undetected until the tokens were already circulating on decentralized exchanges. Neither design flaw required novel exploit techniques, which is often the more damning finding. Audits focus on complex logic bugs, while the losses in this cycle keep coming from mundane key management and missing modifiers.
Neither project has announced a reimbursement plan or a full post-mortem at the time of writing. Fetch.ai said it was investigating the key leak and reviewing converter permissions. NuNet has not issued a public statement beyond confirming the mint was unauthorized. Holders of both tokens are watching for whether the teams pause the affected contracts, since the converter remains live on Ethereum and the compromised deployer key on NuNet has not been publicly rotated.
For traders, the episode is a reminder that AI branding carries no security premium. The market treated the two incidents differently, with FET absorbing a routine drawdown while NTX lost most of its value in hours, but both losses trace back to the same root cause: one wallet, two weak permission designs, and no rate limits standing in the way. Until AI-token projects start treating deployer keys and bridge contracts with the same care they put into tokenomics, security firms expect more September-style weeks.