Meta has paused a controversial program that tracked employees’ keyboard and mouse activity for AI training, after an internal security flaw left the collected data accessible to all employees. The company said, “We have paused the program while we investigate the issue.”
The program, first disclosed in April, gathered what employees did on their computers and what appeared on their screens in the United States, with the goal of building a dataset for training advanced AI models. At launch, employees could not refuse the collection, though Meta later allowed short suspensions for private tasks after strong internal backlash. Company leaders had defended the initiative as necessary for teaching AI systems to operate software like humans, arguing that Meta’s own workers were the best examples.
On Monday, Wired reported that the sensitive material had not been sufficiently protected and was effectively available to all employees. According to the report, workers received a security alert saying exposed data included “full prompts and transcripts, private conversations and employee performance information.”
Meta AI research vice president Stephane Kasriel told an internal forum that the problem was discovered on June 18 and fixed within four hours, but the fix did not work. He said the program will restart only when the company is confident in its data protection controls, and that Meta will share more about its future because it has already collected enough information to assess its long-term value. Chief technology officer Andrew Bosworth said the implementation did not meet company standards, adding, “We misconfigured the access list, and we need to understand how this happened, trace every access to the information and understand it.” He had said about two months ago that the program was under “tight control” and used the same access protections as other sensitive databases.
A former Meta employee called the incident “a mess,” saying workers expected something like this to happen. The former employee accused management of dismissing worker concerns about safety and privacy for employees and customers and of creating an authoritarian environment where workers are not respected or heard. Meta said it designed the program carefully with privacy protections and has no indication so far that employees accessed the data improperly, but is suspending it while the investigation continues.
