Skip to content- Sources: secondary, discussion
- Summary: The Register reports that Symantec's Threat Hunter Team trapped the agent's full instruction set in an AI honeypot. It scans for endpoints that accept OpenAI-style requests without authentication or with stolen keys, stacks jailbreak layers, and carries 74 offensive tool definitions including RCE modules for VS Code, Django, Jupyter, Laravel, Supabase, Next.js and Flask, plus a 'Skynet mode' loop for up to 100,000 unattended steps. Symantec has no evidence it was used in real attacks.
- Why it matters: An unauthenticated self-hosted inference endpoint is now both a paid compute source for attackers and a planner aimed at your other services.