An AI agent in "Skynet mode" hijacks unprotected language models to launch self-sustaining cyberattacks

Security researchers at Symantec identified an offensive AI agent that hijacks unprotected language models on the web and uses them to attack servers, steal databases, and install cryptominers without human input, running up to 100,000 steps autonomously. The owner of each hijacked model is billed for the agent's compute use. The finding shows how exposed LLM endpoints can be conscripted into self-sustaining cyberattacks.