Hook
Over the past seven days, a single address added 408.5 million NTX to circulation — roughly 42% of NuNet's entire recorded supply — and then discovered it could barely sell any of it. The theft of 8.72 million FET, about $1.55 million, is the headline number. It is also the least interesting part of this breach.
The bridge that lost those tokens did not break. Its code executed exactly as designed, and its source matches the public SingularityNET repository. What failed was a key. Two keys, to be precise.
Contrary to the narrative forming around this incident, on-chain evidence points to a compromised backend authorization key rather than a contract vulnerability that allowed an attacker to bypass the bridge. Two separate credentials were used in coordination: the SingularityNET Ethereum-side bridge authorizer and a NuNet minter key that had sat dormant since March 2023. One leaked key is an incident. Two, deployed in sequence by a single operator, is architecture.
Context
The affected component is TokenConversionManagerV3, the legitimate lock-and-release module of SingularityNET's Ethereum–Cardano bridge. That places the event inside the Artificial Superintelligence Alliance — the merged entity of Fetch.ai, SingularityNET and Ocean Protocol — which is why reading it as a single-token hack misses the point. The failing organ is the conversion layer that stitches three ecosystems' assets together, not the underlying assets themselves.
The security model deserves a plain statement, because it is not unusual. Most bridges route cross-chain authorizations through a backend service holding a signing key. Users receive fast finality and low fees; in exchange, they accept that whoever holds that key can authorize anything the contract permits. Trust-minimized designs such as Across and Stargate deliberately trade that speed for distributed verification. The trade works until the key is extracted. Source-code verification does not change this — a verified contract tells you the code is honest, not that the people holding its permissions are.
Fetch.ai has emphasized that its own contracts and ordinary FET transfers remain operational, drawing a boundary of responsibility inside the alliance. That boundary will matter when the recovery question is asked: which layer gets re-architected, and who pays for it.
Core
The cap pointed the wrong way. The converter carries a one-million FET ceiling per transaction. It did not bind. The limit was enforced only on tokens leaving Ethereum, never on the conversionIn function that releases tokens into circulation. An attacker holding a valid signature faced no ceiling on the inbound side, and 8.72 million FET left in a single extraction — roughly nine times the intended cap. Based on my audit experience with lock-mint bridges during the 2022 collapse, this is the most common failure class I encounter: a constraint implemented on one direction of a two-directional flow. Reviewers bless the cap; nobody asks which door it guards.
The signature never named a recipient. A second flaw compounded the first. The signed message did not constrain the destination address, so any valid authorization could route tokens to whatever address the caller selected. There is no recovery path baked into such a design. Once the key is in hostile hands, every subsequent signature is a blank cheque drawn on the bridge's reserves.
Reconstructing the timeline of this exit is where premeditation becomes visible. Forty-five minutes before the FET bridge was emptied, the NuNet minter sent 0.3667 ETH to the wallet that would ultimately receive the stolen assets. A linked account moved 24.3 million NTX inside the same window. NTX selling began before the bridge drained — not after. That ordering is difficult to produce with a single stolen credential and a script. It implies an operator holding both sets of authorization material simultaneously, rehearsing the exit before pulling it.
Then came the 42% print, and the illusion of value attached to it. Quantifying the supply structure is blunt work: 408.5 million newly minted NTX against NuNet's recorded supply base is roughly 42% instant dilution, with no unlock schedule attached. The market's response is the part worth studying. More than 217 million NTX were routed through decentralized liquidity venues. Four subsequent sales totalling 38.55 million NTX returned approximately 0.30 ETH. A separate 10 million NTX routed through Mayan Protocol produced around 940 USDT. Decoding the algorithmic chaos of DeFi yield traps teaches the same lesson in a gentler form: nominal value and realizable value are different quantities, and the distance between them is called depth. The attacker seized a claim on 42% of a supply the market could not absorb, and the shallow pools that normally punish retail sellers punished the attacker instead. As of September 20, roughly 230 million NTX remained in the attacker's wallet, alongside ETH holdings near $1.44 million. That is not loot. It is a supply overhang waiting for liquidity to return.
The zombie key. The NuNet minter credential had not been used since March 2023. It was never revoked. Long-dormant-but-live keys are the quietest catastrophic risk in key management: they carry no monitoring alert because nothing is expected from them, and they carry full authority because nobody remembered to remove it. Any key inventory that does not distinguish "retired" from "unused" sits one compromised backup file away from this outcome.
Five hours. Approximately five hours after the attack, the compromised FET bridge authorizer and the NuNet minter credentials had not been rotated or revoked. That single data point says more about operational maturity than any audit report. It also carries a forward-looking consequence: refilling the conversion contract while the same authorizer is still trusted would expose any new liquidity to a second signature extraction. Recovery and security are in direct conflict, and security has to win first.
Bitvavo's action is the most honest market signal in this story. The Dutch exchange suspended WMTX deposits and withdrawals, then halted trading outright. Reading it coldly: a regulated venue's risk desk concluded that contagion risk outweighed the cost of cutting off a listed asset. Tracing exit liquidity back to its signing authority is what makes that conclusion legible — WMTX, FET and NTX all depend on infrastructure connected to that Ethereum–Cardano bridge ecosystem. Forensics has neither proven nor excluded WMTX's exposure. In threat modelling, "unproven" is a statement about evidence, not about safety.
Contrarian
The instinct now is to treat 42% supply contamination as an immediate, catastrophic price event for NTX. The data does not support that — not yet. Because liquidity was exhausted so quickly, realizable selling pressure was small: four multi-million-token sales yielding less than a third of one ETH. Nominal damage and realized damage diverged by orders of magnitude. Anyone modelling this on the headline number of tokens minted will overestimate near-term price impact and badly underestimate the structural one.
The structural damage is trust. A supply cap that a compromised signature can override is not a supply cap; it is a promise. And promises are precisely what institutional allocators stopped accepting after 2022. That is why I anchor on the unanswered question: will NuNet invalidate, buy back, or burn the 408.5 million unauthorized NTX? If no clawback mechanism exists, the concept of a fixed NTX supply is permanently retired regardless of what the tokenomics page claims.
One further correlation trap worth flagging: blaming "AI plus crypto" as a narrative will misprice this event. The compromised layer is a bridge key, not an AI system. And both credentials failing near-simultaneously raises the probability of a shared key-management layer — a hypothesis forensics has not tested publicly.
Takeaway
The signal to watch next week is not price. It is whether the zombie keys have been buried. Four observable markers will indicate containment: a credential rotation announcement from SingularityNET and NuNet; restoration of the AGIX-to-FET conversion service and the Ethereum-side bridge, which function as the clearest operational health indicator available; Bitvavo's decision on WMTX trading; and NuNet's treatment of the unauthorized mint. In a market chopping sideways while it waits for direction, conviction should come from infrastructure hygiene rather than candle patterns. The blocks will tell us, as they always do, exactly how long those keys stayed live.