Mercor · 2026-04-09 · major
Mercor LiteLLM Supply-Chain Breach — 4TB Stolen, Meta Pauses AI Training Contracts
Lapsus$ claimed 4TB from AI recruiter Mercor via a poisoned LiteLLM PyPI package — 40k voice samples, video interviews, and Anthropic/OpenAI/Meta training docs. Meta paused all contracts. Five lawsuits filed in one week.
A poisoned open-source AI library package led to one of the biggest AI data breaches of 2026.
What is it?
Mercor, a $10B AI recruiting startup whose customers include Anthropic, OpenAI, and Meta, confirmed a major data breach tied to a supply-chain attack on the open-source LiteLLM project.
How does it work?
Hacking group TeamPCP first compromised Trivy (a security scanner), using stolen credentials to access a LiteLLM maintainer's PyPI account. They then published malicious versions 1.82.7 and 1.82.8 to PyPI on March 27. Any project that auto-updated absorbed the backdoor. Mercor used LiteLLM as its core AI routing layer, giving attackers a foothold.
Why does it matter?
Lapsus$ claimed 4TB of data including 939GB of source code, a 211GB user database, ~3TB of video interview recordings, and ID verification docs for 40,000+ contractors. Training methodology documents belonging to Meta, OpenAI, and Anthropic were potentially exposed. Meta immediately paused all data-labeling work with Mercor. Five federal lawsuits were filed within one week.
Who is it for?
Any team using LiteLLM versions 1.82.7 or 1.82.8 should audit logs and rotate all LLM provider API keys immediately.