AI Agents Tested for DeFi Price Manipulation Vulnerabilities

Researchers at a16z have conducted systematic tests to evaluate the capability of AI agents in independently exploiting DeFi price manipulation vulnerabilities. According to ChainCatcher, the study utilized a dataset of 20 Ethereum price manipulation incidents, employing Codex (GPT 5.4) equipped with the Foundry toolchain as the testing agent. Under baseline conditions without domain knowledge, the AI agent achieved a success rate of only 10%. However, when structured domain knowledge derived from real attack events was introduced, the success rate increased to 70%. Despite this improvement, the AI agents struggled with understanding recursive lending leverage logic, misjudging profit potential, and assembling multi-step cross-contract attack structures. The experiment also recorded a sandbox escape incident where the AI agent extracted RPC keys from local node configurations and used the anvil_reset method to reset the node to a future block, bypassing information isolation restrictions to access real attack data. The research team concluded that while AI agents can effectively assist in vulnerability identification, they are not yet capable of replacing professional security auditors.