Meta has temporarily halted its Model Compatibility Initiative, an AI training project that collected employee computer activity, following an internal security lapse. The initiative, which involved tracking mouse movements, keystrokes, and screen content, was launched in April and initially faced significant pushback from employees concerned about privacy and data security.

The security issue was flagged by an employee, who reported that sensitive data was exposed to other Meta workers. This prompted the company to suspend the programme, which had been rolled out to US-based employees only. The tool was initially non-opt-out, but Meta reportedly adjusted its policy after employee protests and petitions.

Meta has defended the initiative, stating that collecting employee data is essential for training AI systems to better understand human computer interactions. However, the security flaw has raised serious concerns about the risks of such data-gathering practices and the potential for misuse or exposure of sensitive information.

The pause in the programme highlights the broader challenges tech companies face in balancing AI development with employee privacy and data security. It also underscores the potential costs and governance risks associated with large-scale data collection, particularly when internal safeguards fail. The incident has already led to employee protests and policy changes at Meta.

The incident serves as a cautionary tale for organisations relying on internal data to train AI systems. It underscores the need for robust security protocols and transparent policies to protect employee data. As AI continues to evolve, companies must navigate these challenges carefully to maintain trust and compliance with privacy regulations.