CertiK Report: AI Agents Shift to Autonomous Execution in Web3 Security Field

CertiK has released an Intel3D report stating that AI agents are shifting from auxiliary analysis to autonomous execution, capable of autonomously reasoning, calling tools, and taking actions. In the Web3 field, AI agents have been used for contract grading, transaction risk scoring, and tracking stolen funds. The report points out that lightning loans have a fast attack speed, rapid transfer of stolen funds, coupled with a shortage of compliance talents and anti money laundering fines exceeding $900 million in the first half of 2025, promoting the acceleration of security automation. CertiK warns of error output, weakening of manual review, and the risk of attacks against AI. It is recommended to retain behavior records, limit autonomous decision-making boundaries, conduct adversarial testing, and designate responsible persons for each intelligent agent.