NOISESaved copy
Coverage history

Assembled by NOISE.

Saved report copy

Recorded

Autonomous attack agent hijacks exposed AI models, Symantec says

Symantec researchers captured an autonomous attack agent designed to hijack exposed AI models and run up to 100,000 steps without human approval. The agent scanned the internet for unauthenticated OpenAI-style endpoints, then used 74 offensive tools to steal data, execute code, and mine cryptocurrency. Symantec said it found no evidence of real-world attacks, but the kit worked against a live target. The discovery shows attackers now use AI models as reasoning engines inside offensive tools, putting exposed AI systems at risk.