Meta has “indefinitely” paused all work with AI recruiting startup Mercor after a breach that attackers claim exposed several terabytes of data.

Two sources confirmed the news to WIRED, adding that the pause is indefinite. Contractors who depended on those Meta projects cannot log hours until, or if, they resume, which could effectively mean they’re out of work. Internal conversations reviewed by WIRED suggest that the company is looking for additional projects for those affected.

Mercor contractors have reportedly not been told why their Meta projects were being paused.

Several other AI labs are also re-evaluating their relationship with Mercor as it investigates the incident, said people familiar with the matter.

The $10 billion AI startup Mercor supplies major AI companies like OpenAI, Meta, and Anthropic with specialized contractors to train and evaluate AI models. However, details about the specific projects and tasks involved are rarely disclosed amid heightened competition between tech giants.

A spokesperson told WIRED that OpenAI is investigating how its proprietary training data may have been exposed in the breach, adding that no user data has been affected. OpenAI has not paused its projects with Mercor.

On March 31st, Mercor confirmed the breach in a staff email: “There was a recent security incident that affected our systems along with thousands of other organizations worldwide.”

The company said it was impacted by a supply chain attack involving LiteLLM, a popular Python library used by AI developers, which was recently infected with credential harvesting malware. An attacker known as TeamPCP took credit for the breach, alleging it accessed 300GB of data from over 500,000 compromised systems.

Lapsus$ hacker group claimed the Mercor attack, saying they accessed four terabytes of the company’s data, including an unnamed database weighing over 200GB and a 3TB drive containing video and verification data.

Our research team attempted to investigate attacker claims, but at the time, the data was inaccessible.

The incident hints at a potential collaboration between TeamPCP and Lapsus$. TeamPCP earlier said it will partner with major illicit forums and ransomware gangs, planning to send invites to over 300,000 registered forum users to become ransomware affiliates.

---

Unlock more exclusive Cybernews content on YouTube.