Meta internally exposed data from its employee-tracking program, which included full prompts and private conversations, according to sources cited by Wired. The program, designed to collect workers' keystroke data to train AI models, saw vulnerabilities that allowed employees to access each other's information.
Employees had previously raised concerns about the initiative, questioning its privacy implications. The blast revealed how even internally focused data-collection efforts can pose risks when safeguards fail.
The exposed data encompassed the full prompts used in the program, as well as private conversations among workers. The incident underscores challenges in balancing AI model training needs with employee privacy protections.
This breach could further erode trust between Meta's workforce and leadership. The programme's future remains uncertain as internal reviews are likely underway.
Some workers may argue the benefits of using real-world data for AI training outweigh privacy risks if properly managed. Meta has not commented on potential policy changes.