Meta paused AI training on employee keystrokes after data was 'put in a place it wasn't supposed to go:' CTO

In a recent interview, Meta's Chief Technology Officer Andrew Bosworth revealed the reasons behind the company's decision to suspend its contentious Model Capability Initiative. This initiative, which aimed to enhance AI training by tracking employee keystrokes and mouse movements, faced significant backlash from staff and was put on hold following a data mishap. Bosworth shared insights with The Atlantic's CEO Nicholas Thompson, explaining that while the data collected was deemed "quite secure," it had been mistakenly relocated by a researcher to an unauthorized area within Meta's systems. He clarified that there was no security breach or foul play involved. "One of the researchers... had put it in a place it wasn't supposed to go," Bosworth stated, emphasizing that the incident was an internal error rather than a malicious act. The Model Capability Initiative, launched in April, required most of Meta's U.S. employees to participate, leading to widespread discontent due to the lack of an opt-out option. Bosworth acknowledged that employee morale was at a low point, possibly the worst in Meta's 20-year history. Following the data leak that made sensitive information accessible internally, the program was paused in June, prompting Meta to initiate a thorough investigation. A spokesperson for Meta confirmed the company's commitment to privacy safeguards, noting that there was no evidence of improper data access by employees. However, the need for a pause was clear as the program was not producing the diverse data necessary for effective AI training. Bosworth explained that the initiative was yielding repetitive data rather than the varied insights required, leading to the decision to introduce expanded opt-out options soon after the launch. He humorously remarked, "A pause, infinite pause. Whenever you don't want to have it, just press pause." The situation continues to unfold as Meta reassesses its approach to AI training and employee privacy.