Meta Halts Mercor Work After AI Training Data Breach

Meta has indefinitely paused all work with data contracting firm Mercor following a major security breach, with other AI labs including OpenAI also investigating the incident. Mercor supplies proprietary training datasets to top AI companies, making the breach potentially damaging as the data could reveal competitive model-training secrets to rivals. The breach is linked to a compromised AI tool called LiteLLM, with a hacker group known as TeamPCP believed responsible. A separate group claiming to be Lapsus$ has offered alleged Mercor data for sale online, including over 200GB of databases and nearly 1TB of source code. Thousands of other organizations may also be affected.