CertiK reports AI agents are evolving from support tools to autonomous cybersecurity operators in Web3, capable of contract triage and transaction risk scoring, though new risks like reduced human oversight require stronger controls.
AI & Agents ·
AI agents are shifting from passive support functions to autonomous operators capable of executing security responses with minimal human oversight, according to a CertiK report. In Web3 environments specifically, these systems now handle contract triage and transaction risk scoring. Within security operations, agents can investigate suspicious activities—such as checking device records and threat feeds following an unusual login—and take direct action like account suspension before flagging results for human review.
The move toward automation stems from operational pressure: flash-loan exploits execute within seconds, stolen assets traverse bridges and mixers within hours, and cybersecurity staffing shortages compound rising regulatory demands. Enforcement consequences underscore the stakes—over $900 million in AML penalties accumulated during the first half of 2025. As agents expand into financial and crypto compliance tasks including exploit detection and fund tracing, their decisions carry direct accountability.
Yet the shift introduces substantive risks. Incorrect agent outputs, diminished human scrutiny, and potential adversarial attacks on AI systems all pose control challenges. CertiK recommends that organizations establish clear decision boundaries for autonomous systems, maintain detailed logs of agent inputs and actions, conduct adversarial testing, and designate explicit ownership for each deployed agent. Without such guardrails, the report warns, automation may simply substitute familiar problems with failures that prove harder to investigate and explain.