CertiK published its Intel3D report on October 5, 2026, detailing the use of autonomous AI agents for blockchain security. These agents are designed to monitor on-chain transactions, track stolen funds across different chains, and automatically pause vulnerable smart contracts within a single block window.
The report highlights the limitations of manual tracing when dealing with complex laundering methods, such as the use of mixers and bridges. AI agents can also generate regulatory reports and activate circuit breakers to halt trading during security incidents, similar to mechanisms used in traditional stock exchanges.
CertiK cited the Bybit exploit as a case study, noting that 86.29% of stolen ETH was converted to Bitcoin within one month. The firm also referenced major breaches at Bitget and the Liquid Network in September 2026 to emphasize the need for rapid automated responses to cyberattacks.
Despite the automation, CertiK stresses that human oversight remains mandatory. Organizations must assign a human owner to each agent, who bears full accountability for the AI's decisions. This is critical because agents may incorrectly freeze legitimate transactions or generate inaccurate reports.
To mitigate risks, including adversarial attacks where bad actors attempt to manipulate the AI, CertiK recommends implementing detailed audit trails, rigorous testing protocols, and capped autonomy for all agents. The report focuses on the broader Web3 ecosystem rather than specific tokens.
