Case mercor-litellm-breachAI supply chainUnited States
Poisoned LiteLLM releases tied to Mercor breach exposing AI lab training secrets
Malicious versions of the LiteLLM gateway library were linked to a breach at AI data vendor Mercor, prompting Meta to pause work amid fears training methods and contractor data leaked.
CriticalAttack on AIStatus: Under investigationVerified case file
Open in the interactive mapCite this case
- Reported
- Apr 3, 2026
- Location
- United States
- Target
- Mercor; clients Meta, OpenAI, Anthropic
- Severity
- Critical
- Triggered by
- TeamPCP (LiteLLM poisoning); Lapsus$ claimed the Mercor data
- Actor type
- Criminal
- Reported loss
- None reported
- CVE
- None
What happened
Malicious versions of the LiteLLM gateway library were linked to a breach at AI data vendor Mercor, prompting Meta to pause work amid fears training methods and contractor data leaked.
Impact: Reported theft of Mercor source code, user database and storage buckets including interview videos, passport scans and SSNs of tens of thousands of contractors; AI lab training workflows feared exposed; Meta contractors put on pause
Who triggered it
Actor: TeamPCP (LiteLLM poisoning); Lapsus$ claimed the Mercor data
Attribution: TeamPCP linked by security researchers (Wiz and others); Lapsus$ self-claimed Mercor access; link between the two groups unconfirmed
Motive: Credential theft, data theft and extortion; stolen data listed for sale
How it happened
Attack vector: Maintainer credentials stolen via Trivy supply-chain attack used to publish trojanized LiteLLM 1.82.7 and 1.82.8 to PyPI; Mercor pulled the poisoned package into its environment
Root cause: Compromised maintainer credentials on PyPI with no secondary release gate, plus downstream consumers auto-installing new library versions without pinning or verification
Timeline
- TeamPCP publishes malicious LiteLLM 1.82.7 and 1.82.8 to PyPI; live about 40 minutes
- Mercor confirms it was among thousands of organizations hit; Lapsus$ claims its data
- Class action filed on behalf of affected contractors; further suits follow
- WIRED reports Meta indefinitely pauses work with Mercor; OpenAI investigates
Response
Mercor contained the incident and hired third-party forensics; Meta paused work, OpenAI opened a review; LiteLLM yanked the versions; multiple contractor lawsuits filed
Current status: Under investigation
Sources (open in a new tab)
- WIRED wired.com
- TechRepublic techrepublic.com
- The Next Web thenextweb.com
- Storyboard18 storyboard18.com
Cite this case
Poisoned LiteLLM releases tied to Mercor breach exposing AI lab training secrets. AI Threat Watch, case mercor-litellm-breach, Apr 3, 2026. https://aithreat.io/cases/mercor-litellm-breach/
Permanent link: https://aithreat.io/cases/mercor-litellm-breach/