An Open-Source AI Tool That No One Saw Predicted Kelp DAO's $292 Million Vulnerability 12 Days Ago
An open-source AI security tool flagged critical risks in Kelp DAO’s cross-chain architecture 12 days before a $292 million exploit on April 18, 2026—the largest DeFi incident of the year. The vulnerability was not in the smart contracts but in the configuration of LayerZero’s cross-chain bridge: a 1-of-1 Decentralized Verifier Network (DVN) setup allowed an attacker to forge cross-chain messages with a single compromised node.
The tool, which performs AI-assisted architectural risk assessments using public data, identified several unremediated risks, including opaque DVN configuration, single-point-of-failure across 16 chains, unverified cross-chain governance controls, and similarities to historical bridge attacks like Ronin and Harmony. It also noted the absence of an insurance pool, which amplified losses as Aave and other protocols absorbed nearly $300M in bad debt.
The attack unfolded over 46 minutes: the attacker minted 116,500 rsETH on Ethereum via a fraudulent message, used it as collateral to borrow WETH on lending platforms, and laundered funds through Tornado Cash. While an emergency pause prevented two subsequent attacks worth ~$200M, the damage was severe.
The tool’s report, committed to GitHub on April 6, scored Kelp DAO a medium-risk 72/100—later acknowledged as too lenient. It failed to query on-chain DVN configurations or initiate private disclosure, highlighting gaps in current DeFi security approaches that focus on code audits but miss config-level and governance risks. The incident underscores the need for independent, AI-powered risk assessment tools that evaluate protocol architecture, not just code.
marsbit31m ago
