Meta pauses an AI training program that tracks employees' keystrokes after a companywide leak

Meta has decided to suspend an internal AI training initiative after sensitive employee information became inadvertently accessible. Screenshots obtained by Business Insider revealed that this breach included private conversations, performance metrics, and transcriptions of employees. The severity of the incident has been rated as SEV 2 on a scale where 1 represents the highest level of concern. A spokesperson for Meta acknowledged the situation and stated that the company is currently conducting an investigation. "This program was designed with privacy protections in mind, and while we have no evidence that any data was improperly accessed by our employees, we are pausing the initiative as we look into the matter further," the spokesperson explained. The AI training program, known as the Model Capability Initiative (MCI), was launched in April with the aim of enhancing the company’s AI capabilities by utilizing data from employee keystrokes and mouse movements. However, this mandatory program faced significant pushback from staff who were uncomfortable with having their activities monitored. The recent leak has led to growing discontent among employees, with many expressing frustration over the lack of initial data security measures. One employee voiced their anger in an internal discussion, stating, "I am incensed. While there’s no evidence of malicious access, it’s incredibly frustrating that this data wasn’t secured as promised." This incident is just the latest in a series of security challenges for Meta. Recently, a vulnerability in its AI chatbot allowed unauthorized access to multiple Instagram accounts, and a rogue AI agent was implicated in a serious incident earlier this year, according to reports.