CertiK Proposes AI Agents to Bolster Crypto Security With Human Oversight

A new report from CertiK, published on October 5, explores how AI agents could enhance crypto security by working alongside human teams. These agents could monitor transactions for suspicious activity, trace funds across blockchains, and assist in preparing compliance filings. However, humans would retain ultimate responsibility for critical decisions.

The report outlines potential applications, including the ability to pause smart contract functions or activate circuit breakers when suspicious transactions are detected. AI agents could also reconstruct attack paths and track funds through mixers, bridges, and intermediary addresses. Despite these capabilities, the report does not provide any specific examples of AI agents successfully preventing attacks or recovering stolen assets.

For compliance purposes, AI agents could draft Suspicious Activity Reports based on transaction histories, but human compliance officers would still need to review and certify the reports. The report acknowledges risks such as incorrect findings, wrongful freezes, and inaccurate reports, which could raise legal and ethical questions about responsibility.

To mitigate these risks, CertiK recommends safeguards like maintaining audit trails, defining actions requiring human approval, testing agents against adversarial manipulation, and assigning each agent a named human owner. The deploying organization would remain accountable for the agent’s actions. The report does not compare the performance of AI agents with human analysts but focuses on theoretical capabilities and controls.