According to CoinMeta, the report Intel3D published by Certik states that AI intelligent agents are shifting from assisted analysis to autonomous execution, possessing the ability to reason independently, call upon tools, and take action on their own. In the field of Web3, AI intelligent agents have been applied to contract grading, transaction risk scoring, and tracking of stolen funds. The report points out that flash loan attacks are fast and the transfer of stolen funds is rapid. Coupled with a shortage of compliance professionals and anti-money laundering fines exceeding $900 million in the first half of 2025, this has accelerated the need for security automation. Certik warns of the risks of incorrect outputs, weakened manual reviews, and attacks targeting AI. It recommends retaining behavior records, limiting the boundaries of autonomous decision-making, conducting adversarial testing, and assigning a responsible person for each intelligent agent.