Meta investigates internal exposure of employee keystroke data collected for AI training

Meta is investigating an internal security failure after an internal notice revealed that employee keystroke data, mouse clicks, and screen content were accessible to other employees. The exposed data, collected as part of the company’s Model Capability Initiative, includes full prompts, transcriptions, private conversations, and performance data, according to documents reviewed by Wired.

The initiative, which began in April, tracks activity on corporate laptops of Meta’s US employees to train AI models. Tracy Clayton, a Meta spokesperson, confirmed the company is investigating and stated there is no indication the data was improperly accessed, though employees have raised concerns about the program’s privacy safeguards.

An internal security notice indicated that data across 45,000 hive tables—including sensitive employee activity—was exposed. The incident has fueled internal backlash, with over 1,600 employees signing a petition protesting the surveillance program, citing security and regulatory risks, as well as perceived lack of safeguards.

Meta’s chief technology officer, Andrew Bosworth, acknowledged in an internal post that the tracking program’s implementation fell short of the standards outlined in its privacy review. The incident has been marked as closed, according to sources at Meta, though findings from the investigation will be shared with employees.

Employees have criticized the program, with one engineer describing the laptop screen scraping for training data without consent as an invasion of privacy and exploitation. Meta executives, including CEO Mark Zuckerberg, have defended the initiative, arguing it is necessary to train AI systems to emulate human-computer interactions.

The security incident adds to ongoing morale challenges at Meta, where employees have faced mass layoffs, a turbulent reorganization, and an intensified push to develop AI models. Bosworth recently apologized for ‘atrocious’ communication regarding the AI reorganization and promised improvements, including clearer communication and the return of office perks.