Private data leaked after Meta tracks employees’ keystrokes to train AI
Meta has paused an internal program that tracked employees’ clicks and keystrokes to train artificial intelligence. The pause comes after sensitive data gathered by the program, known as the Model Capability Initiative (MCI), was inadvertently exposed within the company.
An internal security notice sent to U.S.-based employees on Monday, as reported by Wired, revealed that data, including “full prompts and transcriptions, private conversations, people and performance data,” was not properly protected.
“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 said.
Teaching AI to interact with computers like humans
As previously reported by Straight Arrow, the program, which also collects screenshots of employees’ computer screens, was introduced in April with the aim of improving the company’s AI models.
Specifically, according to an internal memo at the time, the AI models would be trained “in areas where they struggle to replicate how humans interact with computers, like choosing from dropdown menus and using keyboard shortcuts.”
The MCI received mixed reviews from employees, some of whom saw the initiative as intrusive. In May, more than 1,600 employees signed an internal petition opposing the program, citing, among other concerns, “the potential for breaches and unauthorized disclosure.”
In response to the exposure, one employee, as seen in screenshots obtained by Business Insider, described themselves as “incensed.”
“I don’t see any evidence of malicious access, but the fact that this data wasn’t locked down as originally promised is super frustrating,” the employee said in an internal group chat.
AI training to continue
Despite the exposure, Meta appears determined to continue the program. Sources at the company told Wired that the incident has since been marked as closed, suggesting that Meta views the problem as resolved.
The incident comes just weeks after a similar data exposure related to Meta’s AI.
Earlier this month the company confirmed that hackers were able to trick its AI support chatbot into changing the passwords for 20,225 Instagram profiles. As a result, accounts such as the one for the White House under former President Barack Obama were hijacked.
Round out your reading
- Not red or blue: America’s politically homeless middle.
- Peter Thiel’s ‘Dialog’ network was super-secret. A data leak changed that.
- The novel legal strategy that Taylor Swift and Matthew McConaughey are using to fight AI.
- Illinois balances budget with new $200 million social media tax that tracks in-state users.
- When Trump serves up ‘Just the News,’ it comes with a side of bias.
Ella Rae Greene, Editor In Chief
