Berri AI: AI incident

In March 2026, hacker group TeamPCP compromised LiteLLM, an open-source tool used in AI-driven software development.

Within 40 minutes, manipulated package versions (1.82.7 and 1.82.8) distributed via the Python Package Index exfiltrated credentials from affected systems' memory.

The attack exploited prior vulnerabilities in the Trivy vulnerability scanner. In August 2026, security researchers disclosed 195 terabytes of stolen data: cloud keys, repository tokens, AI provider credentials, and secrets from approximately 434,000 CI/CD pipelines across roughly 2,500 organizations, including companies like Nvidia, AWS, and Volkswagen. Investigation revealed some credentials remained active.