CertiK Says AI Agents Could Support Crypto Security With Human Oversight

AI agents could help crypto security teams monitor transactions, trace funds across blockchains and prepare compliance filings, while people retain responsibility for consequential decisions, a CertiK report published Oct. 5 says.

The report describes agents monitoring pending and confirmed transactions for suspicious activity. In advanced deployments with appropriate controls, a detection could prompt a pause in a supported smart contract function or activate a circuit breaker.

Agents could also help reconstruct attack paths and follow funds through intermediary addresses, mixers and bridges. These are potential uses; the report does not identify a specific deployment that stopped an attack or traced stolen assets.

For compliance work, agents could draft Suspicious Activity Reports using transaction histories and other evidence. Human compliance officers would review the narrative, check its supporting evidence and certify the filing.

The report identifies risks including incorrect findings, wrong freezes, disruptive pauses and inaccurate reports. Such outcomes can raise questions about responsibility, which remain unsettled across jurisdictions.

Recommended safeguards include keeping audit trails, defining which actions require human approval, testing agents against adversarial manipulation and assigning each agent a named human owner. The deploying organization remains accountable for the agent’s actions.

The report provides no performance data comparing agents with human analysts. It describes possible capabilities and controls, rather than a documented operational case.