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

Meta CTO Andrew Bosworth shed new light on the data leak that made the company pause its unpopular Model Capability Initiative.

In an interview with The Atlantic CEO Nicholas Thompson, released on Wednesday, Bosworth spoke about why Meta paused the AI training program that involved tracking employee keystrokes.

The interview was filmed in late June.

Bosworth said that data generated by the training program was “quite secure,” with only a small number of people having access, but it had been erroneously moved by one of Meta’s researchers. “One of the researchers who was working downstream with that data—and there was no breach here— but had put it in a place it wasn’t supposed to go,” the executive told Thompson.

The employee data, in a transformed state, had “landed someplace that it shouldn’t have landed internally,” he said, adding that Meta did not suspect foul play.

The company was “locking the whole thing down” until it could get to the bottom of this incident, Bosworth said.

The Model Capability Initiative was introduced in April.

It involved installing software on the majority of Meta’s US employees to track their keystrokes and mouse movements to train its AI models.

The program — and Meta’s instruction that employees couldn’t opt out of it — drew major backlash from its workforce.

Bosworth himself said, during an internal meeting, that employee morale in the company was “probably one of the worst it’s ever been” in Meta’s two-decade history.

However, the program was paused in June after a leak made sensitive employee data accessible to the entire company, according to screenshots seen by Business Insider. “We have carefully designed this program with privacy safeguards, and while we have no indication at this time that any data was improperly accessed by Meta employees, we’re pausing it while we investigate,” a Meta spokesperson told Business Insider in June.

In the interview, Bosworth also shared another reason the program had not gone to plan.

It was generating a lot of the same data, he said, when ideally, the company should have gotten more varied data that could be used to train its AI. “Variance is far more important than a high volume of the same thing that gets collapsed into one example, basically,” he said to Thompson. “So that was why, a couple of weeks after we initially launched it, we added expanded opt-outs for people who didn’t want to do it,” he said. “A pause, infinite pause.

Whenever you don’t want to have it, just press pause.” Representatives for Meta declined to provide further comment in response to a query from Business Insider.