Mercor Hit by Cyberattack via Compromised LiteLLM Package

AI recruiting startup Mercor has confirmed it was the victim of a cyberattack in which hackers exfiltrated data from its systems by exploiting a compromised version of LiteLLM, a popular open source library used to interface with large language model APIs. The attack is attributed to an extortion-focused hacking crew that leveraged a supply chain compromise — meaning the malicious code was embedded in the dependency itself, not Mercor's own codebase directly.

LiteLLM is widely used across the AI development ecosystem as a unified interface for calling models from OpenAI, Anthropic, Google, and others. A compromise at that layer is particularly dangerous because it sits upstream of many applications, meaning a single poisoned package can affect dozens or hundreds of downstream products and services simultaneously. The incident underscores how the rapid adoption of AI tooling has outpaced security scrutiny of the open source components powering it.

Mercor, which uses AI to match job candidates with employers and has processed significant volumes of applicant data, has not disclosed the full scope of what was stolen. The extortion crew's involvement suggests stolen data may be used as leverage, raising concerns about the sensitivity of the recruitment and personal information potentially exposed. The company says it is notifying affected users and working with security researchers to assess the damage.

This incident joins a growing list of supply chain attacks targeting developer infrastructure — from the SolarWinds breach to the xz Utils backdoor — but signals a new front specifically within the AI toolchain. As AI startups increasingly rely on shared open source middleware to accelerate development, the security posture of those dependencies becomes a critical and often underexamined attack surface.