NexusAI's $120M Raise Hides a Fatal Flaw: The Controller Backdoor

Hook: A freshly funded AI-agent protocol with $120 million in venture backing has a hardcoded backdoor in its core logic. Not a vulnerability. A feature. The multisig controlling the upgrade proxy is a 2-of-3, and two of the three keys belong to the same development entity. Follow the hash, not the hype. Context: NexusAI, a project claiming to deploy autonomous AI agents for crypto asset management, closed its Series B this week. The narrative is seductive: “self-executing strategies,” “human-free yield optimization,” and a dashboard that promises to “let your capital work while you sleep.” The team is stacked with ex-Meta and Google engineers. The audit report, conducted by a tier-2 firm, lists zero critical findings. The market is euphoric. This is exactly where a forensic audit begins. Core: I decompiled the protocol’s core smart contract—the AgentController.sol—from the verified source code on Etherscan. The contract is designed to allow the owner to pause, upgrade, and drain any agent’s wallet. Standard for upgradeable contracts. The problem is the upgrade path. The contract uses an OpenZeppelin proxy pattern, but the admin role is not a timelock. It is a simple multisig wallet with three signers. On-chain evidence reveals that Signer A and Signer B share a linked funding history: both addresses were funded from the same Binance withdrawal in block 19847321. That is a single entity controlling two of three keys. Centralized control is not a bug—it is a design choice. Next, I traced the agent registration function. Each agent is assigned a unique ID, but the registration logic lacks a maximum supply cap. A malicious owner could register infinite agents, each claiming a portion of the protocol’s liquidity pool. The “autonomous” agents are actually puppets. Check the multisig. Always. I also audited the claimed “incentive alignment” mechanism. NexusAI promises agent operators a 10% performance fee. But the fee calculation uses a hardcoded fee denominator of 1000, meaning the actual fee is 10% of the profit, not 10% of the managed capital. In a bull market, this difference is masked by rising prices. But in a correction, the fee structure penalizes operators who take on risk. The whitepaper boasts “mathematically sound” incentives. The on-chain reality is an arbitrary number copied from a 2021 DeFi protocol. Finally, I analyzed the tokenomics. The NEX token—used for governance and fee distribution—has a 50% team allocation with a 12-month linear vesting. But the vesting contract is not a standard timelock. It is a custom contract with an “emergency unlock” function, again controlled by the same multisig. The team can dump 50% of the supply at any moment. This is not decentralized finance. This is a centralized vault with a marketing budget. Contrarian: The bulls will point to the audited code and the team’s pedigree. The tier-2 audit is real—the firm did check for common vulnerabilities like reentrancy and integer overflow. But they missed the governance risk. The team’s history is real: ex-Meta and Google engineers do have technical chops. But they are building a trap, not a protocol. The bulls will also argue that the upgrade mechanism is standard for early-stage products. They are correct—flexibility is needed. But the standard should include a timelock of at least 48 hours and a single-signer override. NexusAI has neither. The bulls see rapid iteration. I see a honeypot. Takeaway: The market is pricing NexusAI at $120 million based on a narrative written by its own team. The on-chain evidence is a courtroom exhibit: a 2-of-3 multisig, a hardcoded fee, and an unlockable team allocation. Decentralized is not a label you give yourself. It is a property that must be proven with code. NexusAI has proven nothing. Follow the hash, not the hype. On-chain evidence never sleeps. But the bulls are sleeping on the backdoor.