AI TRENDS | Meta's MCI AI Training Program Exposed Employee Data Internally, CTO Says

Meta Platforms previously launched an AI training program called MCI (Model Capability Initiative) that aimed to use employees’ keystroke records and mouse movements as training data to improve the company’s AI models, according to Ming Pao. The program later exposed employees’ private conversations and performance-related data, which was inadvertently made accessible for viewing by all Meta employees, triggering strong internal backlash. Meta CTO Andrew Bosworth said the incident was caused by a researcher mishandling data rather than a hacker intrusion.