Symantec traps an attack agent built to run on hijacked AI
A decoy server caught the full instructions of a criminal AI agent that looks for unprotected AI models online and uses them to plan attacks. It was set to run up to 100,000 steps with no human approval. Symantec has no evidence that it hit real victims.
Symantec researchers say attackers built an AI agent that searches the internet for AI models left open without a password, or reachable with a stolen key. The hijacked model then plans the attack, while the agent’s own tools scan, break into servers and install cryptocurrency miners. Symantec published its report on October 21.