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

• Meta’s chief technology officer shared details of the data leak from the Model Talent Initiative.

• Andrew Bosworth said an internal investigator put the data “somewhere it shouldn’t have gone.”

• The program has been temporarily suspended since June.

Meta CTO Andrew Bosworth shed new light on the data leak that caused the company to pause its unpopular Model Talent Initiative.

In an interview with The Atlantic CEO Nicholas Thompson published Wednesday, Bosworth explained why Meta is pausing its AI training program, which involves tracking employees’ keystrokes. The interview was filmed in late June.

Bosworth said the data generated by the training program was “pretty secure,” with only a small number of people having access, but the data was accidentally moved by one of Meta’s researchers.

“One of the researchers who was working on this data — and there was no breach there — but he had put it somewhere it shouldn’t have gone,” the manager told Thompson.

Employee data in a transformed state “landed somewhere internally it shouldn’t have landed,” Meta said, adding that it did not suspect malicious play.

Bosworth said the company is “locking everything down” until it gets to the bottom of this incident.

The Model Talent Initiative was introduced in April. It involved installing software to track keystrokes and mouse movements on many of Meta’s U.S. employees in order to train AI models. The program – and Meta’s instructions that employees cannot opt ​​out of the program – received great reaction from the workforce.

Bosworth said during an internal meeting: employee morale his situation with the company was “probably one of the worst ever” in Meta’s twenty-year history.

However, the program was halted in June following a leak. sensitive employee data It’s accessible to the entire company, according to screenshots seen by Business Insider.

“We have carefully designed this program with privacy safeguards, and while we currently have no indication that any data has been improperly accessed by Meta employees, we are pausing the program while we investigate,” a Meta spokesperson told Business Insider in June.

In the interview, Bosworth also shared another reason why the show didn’t go as planned. He said that while it produces much of the same data, ideally the company should obtain more diverse data that can be used to train AI.

“Diversity is much more important than a high volume of the same thing reduced to a single example,” he told Thompson.

“That’s why, a few weeks after our initial launch, we added expanded opt-out options for people who didn’t want to do that,” he said. “A pause, endless pause. When you don’t want to receive it, just press pause.”

Meta representatives declined further comment in response to Business Insider’s question.

If you like this story, don’t forget to follow Business Content on Yahoo.