← All Technology stories
Technology / 24 Oct 2026

The AI attack agent that hijacks models

Symantec researchers uncovered the full instruction set of an autonomous attack agent that hijacks exposed AI models and can run up to 100,000 steps without human approval.

Watch on YouTube
01 / THE REPORTING

Sources.

Autonomous attack agent in 'Skynet mode' hijacks other models to carry out its evil deedswww.theregister.com
02 / THE VISUALS

Media credits.

Al Nahian (Pexels License)www.pexels.comPexels (Pixabay Content License)pixabay.comNemuel Sereti (Pexels License)www.pexels.comBibek ghosh (Pexels License)www.pexels.com
03 / THE CONTEXT

Context and limitations.

  • Symantec has not attributed the agent to any person or criminal group, and it found no evidence it was used in real attacks. Avoid implying an identified attacker or confirmed victims. The research describes the agent's design from captured instructions; do not present the malicious prompts as usable instructions.