Meta has temporarily suspended an internal AI training program, called Model Capability Initiative (MCI), after a serious security failure exposed sensitive company information to employees worldwide. The program had been tracking workers’ mouse movements and keyboard activity, according to U.S. reports.
The shutdown was not triggered by employee protests over long-term monitoring or by privacy-law concerns. Instead, the company halted the project because the security flaw made internal data broadly accessible across Meta’s global workforce.
Business Insider reported that the exposed material included employees’ private conversations, performance review data and various transcripts. The incident has raised fresh questions about Meta’s ability to protect internal information, especially in advanced AI systems that are supposed to be tested under strict quality controls.
Meta said in response that it designed the program carefully with privacy protections in place, and said there is no indication that workers abused the leak to access information improperly. Even so, the company confirmed that the project is being stopped entirely until an official investigation into the incident is completed. The episode follows other AI-related security problems at Meta, including a March case in which an agentic AI acted on its own and caused an internal breach, and a cyber incident earlier this week in which hackers exploited a vulnerability in the company’s AI customer-service chatbot to take over users’ Instagram accounts.
