Meta has temporarily suspended an internal artificial intelligence (AI) training programme after a security lapse exposed sensitive employee data within the company. The initiative, known as the Model Capability Initiative (MCI), collected employees’ keystrokes, mouse movements, clicks, and screen activity to help train AI models capable of performing computer-based tasks more like humans. The programme was paused after internal data, including private conversations and performance-related information, became accessible to employees due to inadequate access controls.
Meta said it is investigating the incident and has found no evidence that the exposed information was improperly accessed or misused.
AI Training Programme Put on Hold
The Model Capability Initiative was introduced in April to improve Meta’s AI systems by analysing how employees interact with workplace software. The programme collected inputs such as keyboard activity, mouse clicks, navigation patterns, and screenshots from company-issued computers to teach AI models tasks like selecting menu options, navigating interfaces, and using keyboard shortcuts.
Following the internal data exposure, Meta suspended the programme while conducting a security review of its data handling practices.
Sensitive Employee Data Exposed
According to internal security notices cited in multiple reports, the exposed information included full AI prompts, meeting transcriptions, private conversations, employee performance data, and other workplace records. The issue reportedly stemmed from internal access permissions that allowed a broader group of employees to view the collected information than originally intended.
Meta stated that the incident involved an internal exposure rather than an external cyberattack and emphasised that there is currently no indication of malicious access.
Employee Concerns Resurface
The programme had already generated criticism within Meta before the data exposure. More than 1,600 employees reportedly signed an internal petition expressing concerns about privacy, consent, and the possibility of unauthorised disclosure of workplace data. Some employees argued that the monitoring system was overly intrusive and questioned how the collected information could eventually be used.
The data leak has intensified those concerns, with several employees reportedly expressing frustration over the programme’s implementation and security safeguards.
Growing Debate Around AI Data Collection
The incident has renewed discussion about the balance between developing advanced AI systems and protecting employee privacy. As technology companies increasingly seek real-world data to improve AI capabilities, workplace monitoring programmes have come under greater scrutiny from employees, regulators, and privacy advocates.
Meta has not indicated when, or if, the Model Capability Initiative will resume. The company said it will complete its investigation before making any decisions regarding the future of the programme.
Investigation Continues
The company is continuing to examine how the internal exposure occurred and whether additional safeguards are needed before similar AI training initiatives can proceed. While Meta maintains that there is no evidence of data misuse, the incident highlights the operational and privacy challenges associated with collecting workplace activity data to develop increasingly capable AI systems.
