Meta has suspended an internal programme intended to collect data on how employees use computers, following concerns over workplace privacy and the possible use of the data to automate certain tasks.
The programme, known as the Model Capability Initiative (MCI), was developed by Meta’s artificial intelligence division TBD. Its objective was to collect computer-use data that could help researchers develop AI systems capable of carrying out routine tasks on computers.
According to reports, the programme involved collecting information about activities such as keystrokes, mouse movements and screen interactions. Meta’s AI teams viewed this type of employee-generated data as potentially useful for improving AI systems that can already perform specialised activities, including coding and research, but have limitations when handling everyday computer tasks.
The initiative reportedly faced resistance from employees following its internal rollout in April. Workers raised privacy concerns about the possibility of sensitive information being captured during computer activity. Examples cited in employee discussions included passwords, financial information and immigration-related documents.
Some employees also expressed concern that the data could eventually be used to develop AI systems capable of performing work currently handled by employees.
Opposition reportedly included internal campaigns, informational material and a petition signed by more than 1,800 Meta employees. Workers also raised concerns about an opt-out mechanism introduced in June, which allowed employees to temporarily disable the software for personal activities.
The programme encountered another issue after employees reportedly found that data collected through the software could be accessed across Meta’s organisation. Reports also indicated that some private conversations had been captured.
Meta subsequently suspended the initiative and indicated that any future version would operate on an opt-in basis.
The programme was reportedly still suspended two months later. There has also been uncertainty about whether information collected through MCI was used to train Meta’s newer AI models, with differing assessments regarding the data’s use and effectiveness.
The episode highlights the challenges technology companies face when collecting employee-generated data for AI development, particularly around privacy, consent and the potential impact of workplace automation.
Get it on Play Store