LiteLLM 1.107.0 was poisoned: rotate every secret it could reach
A hijacked LiteLLM release pipeline pushed a credential stealer to PyPI and Docker for about two hours on 21 October. Rebuild on 1.106.2 and rotate secrets.
AI Security
Attacks that use AI, and attacks on AI systems, mapped to OWASP and MITRE ATLAS.
Start here: OWASP agentic Top 10 meets MITRE ATLAS: a threat model
AI in the hunt, part 1: Microsoft’s Project Perception and Codename MDASH
Microsoft's Project Perception puts red, blue and green AI agents inside Defender. Codename MDASH hunts bugs in source code. What each does, what hunters can use, and what to check before trusting the output.
OWASP agentic Top 10 meets MITRE ATLAS: a threat model
OWASP says what goes wrong with AI agents; MITRE ATLAS says what attackers do to them. We map one to the other and note what ATLAS added this autumn.
Four AI agent advisories, one lesson: prompt injection becomes code execution
Four October advisories in Flowise, PraisonAI, Vibe-Trading and LangAlpha share one chain: text the agent reads becomes code it runs. Patch, then fix the architecture.
