Meta Suspends Collaboration with Mercor Over Data Breach Risking AI Industry Secrets Exposure

Meta has indefinitely suspended all collaboration with data contractor Mercor in order to investigate a large-scale security breach that occurred at the company. Mercor is a key partner that provides customized datasets for model training to major AI labs including OpenAI and Anthropic.

The incident is believed to have originated from a breach of two versions of the AI API tool LiteLLM by the attacker TeamPCP. As a result, numerous companies and services using LiteLLM may have been affected, and the Mercor security incident points to the risk of exposure of sensitive data related to how AI models are trained.

The hacking group Lapsus$ claimed to be selling Mercor's data, but security experts assess that the attack was likely carried out by TeamPCP. OpenAI is currently investigating the security incident and has confirmed that it did not affect its user data.