Mental health at work has never looked so fragile. In June 2026, the 16th Empreinte Humaine–Ipsos BVA barometer delivered a figure never seen before: 50% of French employees show signs of psychological distress, the highest level measured since 2020, and 83% attribute it at least in part to their work.
At the same time, artificial intelligence is settling into organizations with a promise: freeing us from tedious tasks. The intention is a good one. But on the ground, AI doesn’t always lighten the load; sometimes it adds another layer of mental load, at the worst possible time.
So the question isn’t “for or against AI”. It is: how do you roll it out without it becoming, in turn, a psychosocial risk factor?
Technostress and cognitive overload: the paradox of AI at work
The first paradox: the tool that is supposed to save time also takes time. You have to learn it, write the right instructions, interpret its answers, track down its errors (the famous “hallucinations”), and sometimes redo everything. Two Berkeley researchers who spent eight months in a company that had adopted generative AI describe work that became more complex and faster-paced rather than lighter: constant multitasking, non-stop checking of the output, longer days (Ranganathan & Ye, Harvard Business Review, 2026).
The figures confirm this feeling: 14% of employees say they have already experienced cognitive overload while working with AI: mental fog, flagging concentration, more frequent mistakes, with productivity declining beyond three AI tools used in parallel (franceinfo, 2026). Add context switching on top: constantly jumping from one tool or task to another fragments attention and costs precious minutes of refocusing at every switch.
Finally, a quieter anxiety is rising: not immediate replacement, but a vague feeling of being left behind, of having to “chase” a technology that moves faster than you do. Specialists call it technostress (moka.care, 2026). For people who are already overstretched, all of this directly feeds the risk of burnout.
What are the 4 psychosocial risks linked to artificial intelligence?
Occupational health services keep pointing it out: AI rarely creates new risks; above all, it amplifies existing weaknesses.
- Supervision overload. Checking, correcting and cross-checking AI deliverables, especially when running several agents at once, becomes a task in its own right, and a cognitively costly one.
- Fragmented attention. Constantly juggling between tools breaks up concentration; beyond three AIs, the benefit reverses.
- Loss of meaning and autonomy. When AI decides, prioritizes or writes, the employee can feel like a mere “machine supervisor”, with less room to maneuver and conflicts of values.
- Insecurity. Uncertainty about the future of one’s job and skills keeps a background anxiety going, the core of technostress.
Mental load isn’t the only hidden cost of AI: it also weighs on the environment. Read: what is the impact of AI on the environment?
AI and the French Labor Code: what are the employer’s obligations regarding digital risks?
Many people don’t know it: the framework already exists. The French Labor Code requires employers to take the measures needed to protect the physical and mental health of their employees, which covers the risks created by a new tool or a new organization, and therefore by AI (French Labor Code, articles L.4121-1 to L.4121-5). At the European level, the AI Act strengthens the transparency requirements for AI systems. A poorly governed rollout will sooner or later expose employers to disputes before the labor courts: better to anticipate than to suffer the consequences.
How do you bring AI into the company without harming mental health?
At Teamstarter, we hold a simple conviction: everything starts with letting employees themselves tell apart the uses that relieve them from the ones that overload them. Nobody knows this better than the people who do the work.
Then come moments of exchange between peers, to share practices, tips and limits: what isolates people weakens them, what is shared protects them. The barometer confirms it: without recognition or support from the group, the level of distress climbs sharply (Empreinte Humaine x Ipsos BVA, June 2026). Rolling out AI from the front line, by relying on the group, turns a source of individual anxiety into a matter of collective intelligence.
Conclusion
Governing AI doesn’t mean slowing it down. It means giving it the conditions to last, humanly as well as economically. An AI that exhausts its users produces no lasting gains; an AI that is chosen, shared and mastered does.
The real question isn’t whether AI will transform work, but who decides the terms of that transformation.
FAQ: your questions about AI and mental health at work
Does AI affect employees’ mental health?
Yes, when it is rolled out badly: instead of lightening the work, it can add a layer of mental load. Two Berkeley researchers who spent eight months in a company equipped with generative AI describe work that became more complex and faster-paced (Ranganathan & Ye, Harvard Business Review, 2026), and 14% of employees say they have already experienced cognitive overload while working with AI. This risk comes at a time when 50% of French employees show signs of psychological distress (Empreinte Humaine–Ipsos BVA, June 2026).
What are the psychosocial risks linked to AI?
Four risks stand out: supervision overload (checking and correcting AI deliverables becomes a task in its own right), fragmented attention (beyond three AI tools, the benefit reverses), loss of meaning and autonomy when AI decides or writes in the employee’s place, and insecurity about the future of one’s job, at the core of technostress. AI rarely creates new risks: above all, it amplifies existing weaknesses.
What are the employer’s obligations regarding AI-related risks?
The French Labor Code already requires employers to protect the physical and mental health of their employees, including against the risks created by a new tool or a new organization, and therefore by AI (articles L.4121-1 to L.4121-5). The AI Act, for its part, strengthens the transparency requirements for AI systems. A poorly governed rollout will sooner or later expose the employer to disputes before the labor courts.
How do you roll out AI without harming your teams’ mental health?
Let employees themselves tell apart the uses that relieve them from the ones that overload them: nobody knows this better than the people who do the work. Then organize moments of exchange between peers to share practices, tips and limits, because what isolates people weakens them and what is shared protects them. An AI that is chosen, shared and mastered produces lasting gains; an AI that exhausts its users doesn’t.
