Meta plans to monitor the mouse movements, clicks, and keystrokes of its US employees to create training data for AI agents. The initiative, detailed in internal memos from Meta Superintelligence Labs, targets interactions on work-related apps and websites. A company spokesperson confirmed the data will improve AI handling of computer tasks without evaluating staff performance.
Internal memos from Meta's Superintelligence Labs outline the Model Capability Initiative, a new software that tracks employee actions on specific work apps and sites. The tool also captures periodic screenshots for context. “This is where all Meta employees can help our models get better simply by doing their daily work,” one memo states, as reported by Reuters. The effort focuses on US staff, as similar monitoring in Europe could violate local laws on employee tracking. Meta has previously faced EU scrutiny over AI training data practices requiring users to opt out rather than opt in. Spokesperson Andy Stone explained to Reuters that the data addresses AI weaknesses in tasks like mouse movements, button clicks, and dropdown navigation. “If we’re building agents to help people complete everyday tasks using computers, our models need real examples of how we actually use them,” Stone said. He emphasized the data will not assess employee performance. This approach responds to challenges in sourcing high-quality data for AI interactions, amid a trend where firms like OpenAI, Anthropic, Google, and Perplexity roll out agents that control computers or browsers. Meta has set AI usage targets for some staff and plans workforce reductions of up to 10 percent from May.
