[Report: AI Agents Taking on More Execution Tasks in Cybersecurity] Golden Finance reports on October 6, citing CryptoPotato, that blockchain security company CertiK has released the Intel3D report, which states that AI agents are transitioning from assisting in analysis to executing responses. These agents can autonomously reason, invoke tools, perform forensics, and take action with minimal human intervention. In the Web3 space, they are already being used for contract classification, transaction risk scoring, and tracking stolen funds. The report highlights that flash loan attacks can drain protocols within seconds, with stolen funds being transferred across bridges and mixed within hours. Coupled with a shortage of compliance talent and over $900 million in anti-money laundering fines expected in the first half of 2025, this is accelerating the push for security automation. CertiK warns of risks such as erroneous outputs, weakened human oversight, and attacks targeting AI systems. It recommends maintaining behavioral logs, limiting the scope of autonomous decision-making, conducting adversarial testing, and assigning a responsible individual for each AI agent.