AUTONOMOUS ATTACK AGENT IN 'SKYNET MODE' HIJACKS OTHER MODELS TO CARRY OUT ITS EVIL DEEDS

Summary from the TLDR InfoSec newsletter

Symantec captured an AI attack agent built to find unauthenticated or stolen-key OpenAI-compatible endpoints, then use hijacked models to plan attacks. Its “Skynet mode” allowed up to 100,000 unattended steps and included 74 tools for scanning, exploitation, credential theft, database dumping, remote code execution, and crypto-mining. Researchers found no proof it was used in real attacks, but the kit ran against a live target.