Why 36% of Workers Hide Their AI Use From Employers

More than a third of employees who use artificial intelligence to accelerate their work are deliberately concealing it from management, driven by a rational fear that demonstrating higher productivity will lead directly to workforce reductions rather than rewards.

A survey of 6,000 desk workers across the United States, United Kingdom and Australia, conducted by workplace search platform Glean, found that 36% of those who reported improved output and quality through AI actively hid the extent of their reliance on the technology. The data aligns with broader anxiety documented in a 2024 Microsoft and LinkedIn workforce survey, where more than half of professionals using generative AI for core tasks expressed concern that doing so would signal they were replaceable.

The reluctance to disclose efficiency gains is not a failure of technological literacy. It is a calculated response to corporate incentive structures. When an employee uses a large language model to compress an afternoon-long reporting task into an hour, revealing that shortcut often triggers a managerial recalculation of headcount requirements rather than a bonus or promotion. Silence becomes a defensive strategy against algorithmic displacement.

This dynamic mirrors a century-old labor conflict. Frederick Winslow Taylor documented the phenomenon in his 1911 treatise on scientific management, describing how piece-rate factory workers intentionally restricted their output—a practice he termed “soldiering.” Workers understood that if management observed them producing more items per hour, executives would simply lower the per-item pay rate, forcing laborers to work harder just to maintain their previous income. Taylor himself acknowledged that he would have done the same under those conditions.

Modern knowledge workers face an identical structural trap, but with higher stakes. A factory worker hiding speed protected their wage; a marketing analyst hiding an AI workflow protects their employment status. When managers ask teams to share their best prompts or automation shortcuts during meetings, the request functions as an invitation to engineer one’s own redundancy. Without contractual protections, the time saved by AI is routinely absorbed by expanded responsibilities or used as justification to consolidate roles.

Addressing this friction requires altering the economic terms of disclosure. Norway recently attempted to institutionalize such a shift. On October 2, the Norwegian government signed an agreement with eight employer and union organizations governing AI deployment through 2029. The pact mandates that workers participate in decisions regarding AI integration and stipulates that resulting efficiency gains must benefit employees alongside corporate margins.

The framework builds on a specific Nordic precedent. In the early 1970s, computer scientist Kristen Nygaard partnered with the Norwegian Iron and Metal Workers’ Union to shape how mainframe computers were introduced to factory floors. That collaboration birthed participatory design, a methodology ensuring end-users influence the architecture of the systems they operate. The new AI agreement attempts to apply that philosophy to generative models.

Transplanting this model to the United States faces severe structural barriers. According to 2025 labor statistics, 52.6% of Norwegian employees hold union membership, providing the collective leverage necessary to enforce such agreements. In the U.S., overall union density sits at 10%, dropping to 5.9% in the private sector. American employers can adopt the rhetoric of collaborative AI integration without the binding arbitration mechanisms that make Norwegian commitments enforceable.

For U.S. companies attempting to unlock hidden productivity, voluntary trial frameworks offer a practical alternative. Organizations can implement six-month pilot programs guaranteeing that participating employees will not face pay cuts or termination as a direct result of demonstrated AI efficiencies. Under these arrangements, workers help determine how reclaimed hours are allocated—whether toward paid upskilling, clearing administrative backlogs, or expanding client capacity—before management unilaterally raises performance baselines.

Such guarantees require written documentation and an internal grievance process. While a non-binding pledge cannot prevent a future executive decision to shrink a department, it alters the immediate risk calculus for the individual contributor deciding whether to share a newly discovered prompt.

The International Labour Organization’s most recent global guidance on generative AI in the workplace continues to emphasize that technology transitions require binding social dialogue, warning that unilateral deployment strategies consistently generate worker resistance and suppress measurable productivity gains.