Mercor is an AI startup that provides training data for notable AI companies, and it has just announced a data breach. Allegedly stolen data has been published on a leak site, containing Slack information, internal ticketing data, and videos of conversations between Mercor’s AI systems and contractors. Four terabytes of data have reportedly been stolen, including database records and source code. 

• What happened? This breach is connected to a supply chain attack involving LiteLLM, an open-source library linking applications to AI services. Mercor told Fortune it was “one of thousands of companies” impacted by the LiteLLM supply-chain incident, linked to TeamPCP, a hacking group. The group inserted malicious code into LiteLLM, harvesting credentials and spreading until it was discovered and removed. 
• Who breached Mercor? Extortion hacking group Lapsus$ has claimed responsibility for the breach. While it is not immediately clear how the group accessed the data, some speculate that TeamPCP and Lapsus$ have begun working together. 
• What happens next? Mercor states it moved to contain and remediate the incident upon discovery. Third-party forensics investigations have begun. 

“The Mercor breach exposes a blind spot that’s been hiding in plain sight,” says Eric Schwake, Director of Cybersecurity Strategy at Salt Security. “As enterprises race to connect their data to AI models through proxies and gateways like LiteLLM, they are creating highly sensitive chokepoints that legacy security tools simply weren’t built to see. When one of these middleware layers is compromised, an attacker doesn’t need to trick the AI, they get direct access to raw API keys, unencrypted prompts, and proprietary data flowing underneath. To a legacy WAF, that exfiltration looks identical to a legitimate AI workload. This incident isn’t an outlier. It’s a preview of the new attacker playbook.”