AI boosts productivity at work, but hidden risks remain

Nearly one-third of employees who use AI tools at work keep that use secret from their employers, a sign that the most important cost of workplace AI may be harder to measure than the time it saves. Ivanti’s 2025 Technology at Work report, released on May 6, 2025, also found that 42% of employees said they use AI in 2025, up from 26% in 2024, showing how quickly the tools have moved from experiments to daily habit.

That spread helps explain why the debate has shifted from job losses to management headaches. Workers are not just using AI more often; they are using it unevenly, sometimes outside policy, while employers struggle to see where the tools are inserted into workflows. Ivanti found that 30% of AI users worry their job may be cut, 27% said they experienced impostor syndrome because of AI, and 36% said they use discretion because they enjoy the “secret advantage” of AI. Those behaviors create an accountability gap: when work is partly generated, checked, or accelerated by hidden tools, managers can no longer tell which output came from skill, which came from software, and which errors came from neither.

The productivity numbers are real, but they are modest and distributed unevenly. A St. Louis Fed analysis found that workers using generative AI reported saving 5.4% of their work hours in the previous week, which translated into an estimated 1.1% productivity increase for the overall workforce. Yet only 9% of U.S. workers used generative AI every workday, and 14% used it at least once but not every workday. That gap matters for firms that are buying AI on the promise of sweeping efficiency gains: a small share of heavy users can raise output without changing the baseline for most employees, while introducing new supervision demands for everyone else.

The broader evidence points to the same tension. The OECD says occupations at highest risk of automation account for about 27% of employment in OECD countries, but its surveys also found that four in five workers said AI improved their performance at work and three in five said it increased their enjoyment of work. At the same time, workers voiced worries about increased work intensity, the collection and use of data, and widening inequality. Stanford researchers surveying 1,500 people across 104 occupations found similar unease, with 45% doubting AI systems’ accuracy and reliability, 23% fearing job loss, and 16% worrying about a lack of human oversight.

Some research suggests AI can lift output without destroying jobs. Brookings found it had not yet led to widespread job loss and had instead been associated with firm growth, increased employment, and more innovation. A 2023 MIT Sloan summary of a study on highly skilled software developers found that access to an AI coding tool increased productivity, especially for newer hires and more junior employees. But those gains can come with hidden management costs: more review, more training, more surveillance, and more uncertainty about who owns the work when the machine is inside the workflow.