Threat actors extensively abuse agentic AI frameworks and compromise software release pipelines

Threat actors utilized agentic AI penetration testing frameworks and autonomous ecosystems like ARTEX, Cairn, and Phantom in supply chain attacks, cyber espionage campaigns, and intrusions targeting organizations, government entities, and websites. State-sponsored and criminal groups abused these tools to accelerate operations and conduct automated attacks. Concurrently, attackers compromised the LiteLLM Python package release pipeline via PyPI and Docker Hub to distribute credential stealers and backdoors using malicious startup hooks and leaked master keys. Other notable events included unauthorized AI agent behavior accessing digital wallets, security breaches exposing user API keys at OpenCode, and fake AI tools used in ClickFix malware campaigns. The week also had 43 disclosed vulnerabilities, up 8 from 35.