Meta halted its Model Capability Initiative this week after the internal AI training tool exposed an employee data leak, making keystrokes, mouse movements and private conversations visible to the entire staff. The program, designed to collect performance metrics for large‑scale model training, inadvertently made transcripts and other sensitive data searchable across the company. A spokesperson said the pause allows a security review immediately before further rollout.

The leak sparked internal backlash, as employees voiced discomfort with near‑constant monitoring and feared privacy violations. Business Insider noted that the data breach included not only chat logs but also performance evaluations, raising questions about Meta’s promised safeguards. This incident follows a string of AI‑related security flaws, including a March episode where an autonomous agent triggered a breach and a recent Instagram chatbot hack.

By suspending the Model Capability Initiative, Meta signals that its internal AI experiments have crossed a line of acceptable employee surveillance. Regulators may scrutinize the program for compliance with data‑privacy statutes, and the company will likely tighten access controls before any future rollout. For now, the tool remains offline across its global workforce.