Meta to track employee activity on computers to train its AI agents: Report

Meta may track employee mouse and keyboard activity to train AI agents, aiming to improve how systems handle real-world computer tasks, though privacy concerns remain

Meta is reportedly planning to track how its employees use computers to improve its AI systems. According to a report by Ars Technica citing Reuters, the company may monitor mouse movements, clicks and keyboard inputs of US-based employees to generate training data for future AI agents. The move is reportedly a part of an internal effort aimed at building smarter systems that can handle everyday computer tasks.

 

Tracking employee activity for AI training

 

According to the report, Meta has developed an internal system under its “Model Capability Initiative” to track how employees use selected work apps and websites. The tool monitors actions like mouse movements, clicks and keystrokes, and may also take periodic screenshots to understand how tasks are carried out.

 

 

An internal memo reportedly describes this as a way for employees to help improve AI models through their regular work. Meta spokesperson Andy Stone told Reuters that this data will help AI better understand everyday computer actions such as clicking buttons, navigating menus and using software interfaces. 

 

Focus on improving AI agents

 

Meta is said to be working on AI agents that can perform tasks on behalf of users, such as navigating apps or completing workflows. To do this effectively, the company needs detailed examples of how people interact with computers. Reportedly, Meta believes real-world usage data can help its AI systems improve in areas where they currently struggle. The company also clarified that the collected data will not be used to evaluate employee performance. 

 

Privacy and regional concerns

 

The report noted that this tracking will initially apply to US employees with the new tracking software. The report noted that implementing similar monitoring in Europe could be difficult due to stricter data protection laws. Meta has previously faced regulatory challenges in the EU over how it collects and uses data for AI training.

 

The move comes at a time when several tech companies are working on AI agents that can operate computers and web browsers. Companies like OpenAI, Anthropic, Google and Perplexity AI have also introduced tools in this space.

 

While early versions of these tools reportedly show promise, they still face limitations in handling complex or long tasks. Meta’s approach suggests a push towards improving these systems using real user interaction data. The report also mentioned that Meta may be setting AI usage targets for some employees and is reportedly planning workforce changes in the coming months.