Mercor, an AI-powered hiring and talent assessment platform, disclosed a data breach affecting candidate data stored on its platform. The breach was traced to a compromised dependency in LiteLLM, an open-source Python library widely used by AI application developers to route requests across multiple large language model APIs. The incident highlights how the AI application supply chain introduces security risks beyond the core model provider relationship.
LiteLLM is used by developers to call OpenAI, Anthropic, Google, and other model APIs through a unified interface. Compromising a library at this layer can affect any application that uses it, regardless of which underlying AI provider is involved. Security researchers identified the LiteLLM vector as a target in multiple AI application breaches in 2025 and 2026.
Mercor's platform collects resume data, interview recordings, skills assessments, and in some cases compensation history and references from job candidates. The sensitivity of this data makes AI hiring platforms a high-value target for credential theft and identity fraud operations.
The incident follows a pattern identified by security researchers: AI-enabled applications often integrate numerous third-party libraries and API connectors to function, each of which represents a potential attack surface. Unlike traditional enterprise software with well-defined security perimeters, AI application stacks tend to be assembled quickly from open-source components that may not have received the same security audit attention as proprietary software.
AI application developers building on open-source foundations should treat third-party dependency security as a first-order concern, conducting regular dependency audits and monitoring for known vulnerabilities in the libraries their applications rely on.
Source: AIAAIC Repository -- https://www.aiaaic.org