According to Reuters, Meta will track mouse slides, clicks and keyboard operations of its American staff for high-quality AI data training.
An internal memorandum issued by the Meta Superintelligence Labs team referred to the tracking software as “Model Capability Initiative”, which will operate on specific work-related applications and websites and regularly intercept screenshots to provide context information for AI training. The memo states: “All Meta employees can help our AI model get better through their daily work.” Meta spokesman Andy Stone said that mobile phone training data would help Meta’s AI agents to perform tasks that are sometimes difficult to accomplish, including “mouse moves, click buttons and browsing down the zip menu”. “If we build smarts to help people use computers to do their routines, then our models need real case studies of practical operations such as mouse moves, click buttons and run the drop-down menu. To this end, we have launched an internal tool to capture such input data on specific applications and to help with training models. We have taken security measures to protect sensitive content and ensure that these data are not used for other purposes.”
The plan triggered a huge shock inside Meta. According to Business Insider, Meta’s employees are angry about the program, and the best comments from the inside say, “This makes me very uncomfortable, how do we get out of the program?” with a series of angry expressions. The internal response by Andrew Bosworth, Chief Technical Officer of Meta, that staff members did not have the option to withdraw from the scheme on the company ‘ s distributed computer further exacerbated the contradiction. The use of anonymous employees to describe this practice as “very anti-Utopia” as a complete record of the smallest behaviour deprived the most basic privacy expectations at work. Many staff fear that providing their own hands-on data-trained AI will eventually replace their jobs. This resistance is even stronger, especially in the context of the large-scale layoffs that Meta has undertaken this year. Similar operations in Europe compared to United States Meta employees will be subject to software monitoring and may violate a number of national laws that restrict employers ‘ follow-up. Meta had previously been reviewed by the EU for the collection of user data for AI training, and the regulator had requested Meta to provide a clear objection form. The privacy organization, NOYB, criticized Meta’s exit process for being too subtle and complex, arguing that Meta should require users to “check for consent options” on the basis of GDP R, rather than having users apply for exit by default after collection.
The Internet contains a large number of texts, images and videos that can be used to train the generated AI model, but it is difficult to obtain high-quality training data for physical action or virtual computer interaction because of important and controversial legal constraints. Some companies have shifted to sophisticated physical simulation techniques, such as sophisticated hand-tracing prostheses, to create human interactive data that the AI robotic model can understand. Meta is doing this at a time when major technology companies, including OpenAI, Anthropic, Google and Perplexity, have recently launched new tools that will enable AI agents to take over the user’s computer or browser to perform specific tasks. Mark Zuckerberg recently pledged to increase his input into the AI project this year and tried to position the company at the forefront of that technology. Meta plans to invest approximately $140 billion in artificial intelligence in 2026, almost double its investments in this area in 2025.
Meta has also reportedly begun to set targets for the use of AI among certain employees, including programmers and engineers. It has also been reported that Meta has been retrenching approximately 2,000 staff in several rounds this year and plans to reduce by up to 10 per cent of the global workforce starting in May.
