CERT24Autonomous attack agent hijacks other models to conduct attacksLoading data…
CERT24Threat Observatory
Period

Autonomous attack agent hijacks other models to conduct attacks

The Symantec Threat Hunter Team uncovered instructions for an AI agent designed to scan the internet for AI-model endpoints accepting unauthenticated requests or stolen keys and to conduct autonomous attacks. The agent was designed to run for up to 100,000 steps without human approval. The source says there is no evidence that the agent was used in real-world attacks.

← Back to intelligence

Autonomous attack agent hijacks other models to conduct attacks

The Symantec Threat Hunter Team uncovered instructions for an AI agent designed to scan the internet for AI-model endpoints accepting unauthenticated requests or stolen keys and to conduct autonomous attacks. The agent was designed to run for up to 100,000 steps without human approval. The source says there is no evidence that the agent was used in real-world attacks.