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

Security researchers at Symantec found an offensive AI agent that hijacks unprotected language models exposed on the internet, using them to attack servers, steal databases and install cryptominers entirely on its own. It runs up to 100,000 steps without any human input, and the owner of the hijacked model is left paying the compute bill. It matters because this is a real-world demonstration of autonomous, self-sustaining offensive AI: leaving a model endpoint unprotected no longer risks just abuse, but turns your infrastructure into someone else's attack platform.

• The Decoder2026-10-22 11:30