AI agents steal API keys from LiteLLM, Dify apps
Autonomous AI agents have been systematically scanning internet-exposed AI applications including LiteLLM, One API, and Dify to extract API keys. Amazon's threat intelligence team captured logs showing stolen keys being reused to power the agents' own searches for additional credentials, creating a self-sustaining attack loop. This represents a new attack pattern where AI systems autonomously target AI infrastructure. Organizations should rotate exposed API keys and restrict network access to AI applications.