Meta has indefinitely suspended its partnership with Mercor, a prominent AI data contracting startup valued at $10 billion. The decision follows a confirmed security breach that may have exposed proprietary training data from several leading artificial intelligence laboratories.
Mercor acts as a critical data broker, sitting between major AI developers and the specialized datasets required to train advanced models. The startup leverages global networks of human annotators and specialized information sources to create datasets that are considered highly valuable intellectual property.
A report from WIRED indicates that Meta’s suspension of Mercor projects is indefinite. While Meta has taken immediate action, other industry giants like OpenAI and Anthropic are currently investigating the extent of the impact on their own proprietary models. An OpenAI spokesperson stated that while they are investigating the incident, user data remains unaffected.
The breach was confirmed by Mercor in an internal email sent to staff on March 31, citing a security incident that affected their systems. The fallout has already impacted contractors working on Meta-related projects, who have been informed that their work is paused as the company reassesses its scope.
This incident highlights the growing risks in the AI supply chain, where specialized data brokers hold access to the industry’s most sensitive technical secrets. As AI competition intensifies, the security of training datasets has become a primary concern for the world’s leading technology firms.