Mercor, a prominent AI recruiting startup valued at $10 billion, has confirmed it fell victim to a cyberattack stemming from a compromise of the open-source project LiteLLM. The incident highlights the growing risks of supply chain vulnerabilities in the AI ecosystem, as Mercor was reportedly one of “thousands of companies” impacted by the breach.
The LiteLLM Connection and TeamPCP
The security lapse originated within LiteLLM, a widely utilized open-source library that facilitates interactions between various AI models. Malicious code was discovered in a package associated with the project, which was promptly removed by the LiteLLM team. Despite the quick fix, the library’s massive reach—with millions of daily downloads according to security firm Snyk—created a significant blast radius.
While the initial compromise of the open-source project was linked to a hacking group known as TeamPCP, the extortion group Lapsus$ has since claimed responsibility for specifically targeting Mercor.
Data Exposure and Extortion Claims
Lapsus$ has shared samples of allegedly stolen data on its leak site to back its claims. The leaked material reportedly includes:
- Internal Slack communications and ticketing data.
- Videos purportedly showing interactions between Mercor’s AI systems and contractors.
- Information regarding the specialized domain experts Mercor employs to train models for clients like OpenAI and Anthropic.
Mercor occupies a vital niche in the AI sector, facilitating over $2 million in daily payouts to a global network of scientists, doctors, and lawyers.
Response and Remediation
Mercor spokesperson Heidi Hagberg stated that the company moved quickly to contain the incident and has engaged third-party forensics experts for a thorough investigation. Following the breach, LiteLLM has also overhauled its internal processes, shifting its compliance certifications from Delve to Vanta.
The incident serves as a stark reminder of how even high-growth startups—Mercor recently secured $350 million in a Series C round led by Felicis Ventures—remain susceptible to flaws in the open-source tools that power the modern AI stack.
