You still have your job.
So why does work feel less safe?

You noticed it in the standup first. There was a time you would have said, easily, that you were behind on something, or that you did not yet understand how a new tool fit into your work. Now you hear yourself skip past it. You nod along to a workflow half the room is pretending to follow. You draft an honest message about being stuck, read it back, and delete it before anyone sees.

Your job is not gone. Nobody has threatened it. And yet something at work has changed shape, and the change is hard to point at, because it is not in the org chart. It is in the air of the room. It has become quietly unsafe to be a person who is still learning.

If that is the feeling, you are not imagining it, and it is not a character flaw. New research suggests it is a measurable effect of how AI is arriving at work, and the injury it does is not to your role. It is to something more basic than that.

Can AI harm your mental health even when your job is safe?

Yes, and the mechanism is the surprising part. You can keep your job, your title, and your salary, and still lose the thing that made work bearable: the sense that it is safe to be imperfect there. A 2025 study of 381 employees found that as AI is adopted around people, their psychological safety drops, and depression rises. The threat that hurt them was never the layoff. It was the room going quiet.

That distinction matters, because most of the conversation about AI and work is stuck on one question, will it take my job, and this research is pointing at a different wound entirely. It is not about whether the machine can do your work. It is about what the arrival of the machine does to whether you can still be human while you do yours.

What is psychological safety, and why does AI erode it?

Psychological safety is a plain idea with a clinical name. It is the shared sense in a team that you can speak up, ask a basic question, admit you are unsure, or get something wrong without it being held against you. It is what lets a senior person say "I don't actually know how this works yet" out loud. When it is present, people take the small honest risks that real work is made of. When it goes, they go silent and start performing instead.

AI erodes it by stacking three kinds of uncertainty on top of each other at the same time. There is technological uncertainty, because the tools keep changing under you. There is task uncertainty, because what your job even is keeps shifting as pieces of it get automated or rewritten. And there is social uncertainty, the quiet dread of being the one who looks slow to adapt in a room where everyone is performing fluency they do not really have. Any one of these is manageable. All three, arriving together, is enough to make people stop raising their hand.

The study that connected AI at work to depression

The research comes from a 2025 paper in the journal Humanities and Social Sciences Communications, by Byung-Jik Kim, Min-Jik Kim, and Julak Lee, with the unusually blunt title "The dark side of artificial intelligence adoption." The researchers followed 381 employees across three separate points in time rather than asking everything at once, which lets them see the effect unfold in a sequence instead of catching a single snapshot.

What they found was a clear chain. AI adoption in the organisation had, in their words, "a significant negative impact on psychological safety, which in turn increases levels of depression." The AI did not raise depression directly. It worked through the erosion of safety in between. Statistical testing confirmed that psychological safety was the bridge carrying the harm from one end to the other. Take that bridge away, and the link between AI and low mood largely disappears.

The danger was never the machine learning your job. It was your workplace quietly deciding it was no longer safe to still be learning yours.

There is one more finding worth holding onto, because it is the hopeful part. The same study found that ethical leadership, managers who are fair, honest, and steady about the change, softened the blow. Where leaders named the uncertainty instead of pretending it away, the drop in psychological safety was much smaller. The erosion is not inevitable. It depends heavily on how the change is led.

Why this hits senior professionals hardest

You would think seniority protects you here, and in the narrow sense of job security it often does. But psychological safety is a different exposure, and senior people are strangely thin-skinned to it, because so much of their standing was built on knowing. For years, being the one with the answer was the job. It was the reason you were in the room.

AI inverts that quietly and quickly. Admitting you are still fumbling with a tool a junior colleague already uses fluently can feel like admitting the ground your identity stood on has moved. So the honest question goes unasked. The confusion gets managed in private, late at night, alone with the screen. And that private management of strain, held silently rather than shared, is the exact texture the research keeps tying to depression.

If you are not sure whether your unease is really about your role or really about your safety, it is worth pulling those apart with something other than 3am rumination. The free AI Automation & Relevance Index on this site gives you an honest, unhysterical read on how exposed your actual work is, which is often calmer than the fear had been quoting you. It tends to reveal that the dread was never really about the tasks a model can do.

It is not the AI. It is the silence around it

Here is the reframe the research quietly hands you. If the harm travels through psychological safety rather than the technology itself, then the thing hurting you is not the tool on your screen. It is the fact that a room full of capable people has silently agreed to stop admitting what they do not know. The AI created the pressure. The silence is what turned pressure into strain.

That is oddly good news, because silence is more workable than software. You cannot slow the tools down. You can, in small and specific ways, refuse to be part of the pretending. The senior person who says, plainly and without drama, "I haven't figured this one out yet either," does more for the room's actual safety than any policy, because they give everyone else permission to be honest too. Safety, it turns out, is contagious in both directions.

What actually helps when AI makes work feel unsafe?

Two questions tend to get fused into one heavy feeling, and separating them is most of the relief. The first is real and concrete: given what you actually do, how exposed is your role to automation? That has an answer, and the fear almost always overstates it. The AI Automation & Relevance Index is built to give you that answer in a few minutes, so the dread has a number to argue with instead of a fog.

The second is quieter and more about you than your job: how much invisible load are you carrying from holding all of this alone? The constant keeping-up, the private catching-up, the effort of appearing fluent, all of it draws down a real and limited resource. The Cognitive Load & Bandwidth Index is built to show you how much of your bandwidth this is quietly eating, which is often more than people let themselves notice until it is named.

And then there is the part no tool can do, which is to stop carrying the harder question by yourself. Whether the room has become unsafe, whether your standing really is slipping or only feels like it, whether the version of competence you built is worth defending or worth updating, these are genuinely hard to see clearly from the inside, not because you lack the intelligence but because you are too close to your own working life to be a fair witness to it. That is what a real conversation is for. Not reassurance that AI is fine, and not a pep talk. A second, steadier vantage point on a situation that, from where you sit, has quietly stopped feeling safe.

Common questions about AI, work, and feeling unsafe