On 21 April 2026, Meta announced that it would install tracking software on the work computers of US-based employees. The programme, which Meta calls the Model Capability Initiative, records keystrokes, mouse movements, and periodic screen snapshots across hundreds of websites and applications, including Google, LinkedIn, Wikipedia, GitHub, and Slack. The data will be used to train Meta's AI agents to perform white-collar tasks autonomously.
Meta is not the first technology company to deploy electronic monitoring in the workplace. Productivity tracking software has been a fixture of corporate environments for more than a decade. What is structurally distinct about the Model Capability Initiative is its stated purpose: this is not surveillance designed to assess whether an employee is meeting their targets, but surveillance designed to generate commercial intellectual property.
Employees at Meta have not been offered the opportunity to opt out. The programme applies to US-based staff, where employment law does not require consent for workplace monitoring at this level of specificity. Meta has stated that the data will not be used for performance assessments and that safeguards are in place to protect sensitive content. Whether those assurances are verifiable by the employees affected is a separate matter.
Cornell University labour researchers, commenting on the announcement, identified the dimension that has attracted least attention: compensation. Workers whose keystrokes and screen interactions are captured for AI training are generating commercially valuable data as a byproduct of their working day. They receive no additional payment for that contribution, they have not been asked whether they wish to make it, and their employment agreements did not contemplate it.
In Europe, a programme of this kind would face significant legal obstacles. Under the General Data Protection Regulation, workplace monitoring must be necessary and proportionate to the purpose it serves. Capturing keystrokes, mouse movements, and screen content across hundreds of applications without an opt-out mechanism is unlikely to satisfy the proportionality standard that European data protection authorities would apply. Meta has not announced equivalent plans for employees outside the United States.
If the behavioural patterns of employees navigating their working day are commercially valuable enough to harvest for AI development, at what point does that extraction become a question about what the employment contract is permitted to claim?
The employment relationship, as it has been understood in Anglo-American law, is an exchange: labour for wages, within terms that are set or assumed. Those terms have always had limits, even where those limits were implied rather than written. What a worker thinks, and how a worker navigates information, makes decisions, and interacts with systems, has not traditionally been counted among the assets an employer was entitled to capture and commercialise. The Model Capability Initiative challenges that assumption directly.
Opinion: The Governance Gap
The constitutional significance of Meta's programme does not lie in one company's engineering choices. It lies in what the programme reveals about a governance architecture that was not designed for this situation.
Data captured from employees is not incidental. It is a record of professional cognition, of how people think through tasks, navigate complex systems, and make decisions under the conditions of their working day. When that record is captured without consent, retained, and used to build AI systems deployed commercially, the worker has contributed something of value for which no account has been given.
The absence of an opt-out is a legal position, not an administrative oversight. It reflects the current state of US employment and data rights law: employers are broadly permitted to monitor employees using technology on work equipment, and workers in the United States do not have data rights protections comparable to those that GDPR provides in Europe. The legal architecture that would resist this programme is precisely what is absent in the US workplace.
That absence has a consequence beyond the individuals currently affected. It shapes what other employers will now consider feasible. Each deployment of employee behavioural data for commercial AI training, without legal challenge or regulatory constraint, establishes a precedent: that this is a legitimate extension of the employment relationship. That precedent is set quietly, in the terms of an announcement rather than the language of a contract.
The question this programme poses is not a technical one. It is whether the employment relationship, as currently constituted, was ever meant to include a claim on the cognitive footprint of the people it employs, and if so, by whom and on what terms that claim was agreed.
HumanSafe Intelligence publishes weekly on AI, workplace rights, and the regulatory frameworks shaping both. If this piece was useful, the best thing you can do is forward it to someone who should be reading it.
Sources
Fortune, "Meta will start tracking employees' screens and keystrokes to train AI tools," 21 April 2026; TechCrunch, "Meta will record employees' keystrokes and use it to train its AI models," 21 April 2026; CNBC, "Meta is tracking employee keystrokes on Google, LinkedIn, Wikipedia as part of AI training initiative," 22 April 2026; Euronews, "Meta to track staffs keystrokes and clicks to train its AI, report," 23 April 2026; Cornell Chronicle, "Meta employee tracking raises consent, compensation questions," April 2026
