Meta pauses controversial AI training system following major internal data breach

Meta has paused its internal AI‑training program, the Model Capability Initiative (MCI), after a major internal data leak exposed sensitive employee information to the entire company. The system tracked workers’ keystrokes, mouse movements, conversations, performance data, and transcriptions to train agentic AI models — but due to a security oversight, this highly sensitive data became accessible to all Meta employees.

Meta claims there is “no indication” the data was improperly accessed, but the incident contradicts earlier assurances that the program had strict privacy controls. It also adds to a growing list of AI‑related security failures at the company, including a rogue agentic AI incident in March and hackers exploiting Meta’s AI customer‑service bot to hijack Instagram accounts earlier this month.

The pause does not address broader concerns about Meta’s controversial employee‑monitoring practices or the legality of collecting such data. Instead, the company says it is investigating the leak before resuming the program.

Read the full story on Engadget →