Workers can acquire a fear of being replaced by AI without anything in their own job changing. That is the finding of "Understanding AI anxiety in the workplace: a multi-method study using fear acquisition theory and the technology acceptance model," by Jarosław Grobelny, Mateusz Klakus, Kacper Szymański and Teresa Chirkowska-Smolak, posted to arXiv on 7 July 2026. Across two experiments, the belief that AI operates autonomously was itself a strong threat signal — enough to produce anxiety in people facing no actual risk of losing their post. The effect showed up across companies of different sizes and profiles.
The mechanism the authors borrow is fear acquisition theory: a person can acquire fear from a threat directed at someone else. Reading a news item about a worker displaced by AI, or watching a clip in which an expert says the AI layoffs start soon, is sufficient input. The paper's phrasing for what makes the signal land is worth holding on to — these cues register as threat because they point to an irreversible redistribution of control, an imbalance of power, and a decline in the standing of human labour and human judgement. The anxiety is not primarily about income. It is about who decides.
The second framework is the technology acceptance model, which normally explains adoption: people take up a technology they consider useful and easy to use. Here the same variable runs the other way. An employee who doubts AI applies to his profession is less likely to believe his own job is at stake; an employee who treats the discourse as inflated advertising concludes the job losses will not arrive either. Conviction that the systems are capable produces the opposite: the more powerful the technology looks, the more plausible it looks as a replacement, and the anxiety rises with it.
That pairing sets up a loop the authors describe in five steps:
A worker reads, hears or watches a story about AI replacing people.
Job-loss anxiety stays weak at first.
Other coverage reports rapid AI progress.
Those stories raise the worker's estimate of what AI can do.
Anxiety jumps sharply — the technology looks more capable, and replacement stories keep arriving in the same feed.
The cycle repeats, and the paper's point is that it can terminate in severe replacement anxiety in someone whose position was never in question. It is a psychological loop that is hard to exit.
The study separates three states that get collapsed in ordinary conversation. AI anxiety is general unease about AI in the workplace, not necessarily tied to losing a post — an employee may keep his job and still be managed by a system, and wonder how even-handed its decisions will be. Job-loss anxiety is ordinary employment insecurity, from financial trouble or a reorganisation that abolishes whole classes of role, with no AI involved. AI replacement anxiety is the compound of the two. The third feels different in kind: being cut in a restructuring is legible and familiar, while being replaced by a system reads as a science-fiction scenario with no defence and no exit.
None of which has yet shown up as mass unemployment. Employees are still doing the work they were doing. AI has so far arrived as one more tool a worker is required to use, and current systems have not produced large-scale displacement out of professions. The counter-argument on offer is that this is the lull before the storm — that the serious economic, social and psychological effects land when the systems start doing most of the work. The discussion around that possibility is where universal basic income keeps surfacing, as a floor under everyone regardless of employment status. And the intuition that trades are safe — the plumber and the electrician — holds only until humanoid machines can move, reach the object they need and perform physical tasks at human level, which is the expectation now on the horizon. The office worker was supposed to fall first because AI already handles analytical tasks; the exemption granted to manual work was always conditional on hardware.
Here is what the finding implies and the paper does not say out loud. If perceived AI capability is the variable that converts background noise into personal dread, then the vendors' own capability claims are an input into workforce mental health. Every demo that lands, every benchmark that gets quoted, every launch of ChatGPT, GPT-5, Claude, Gemini, Copilot or Grok that convinces a lay reader the thing is more capable than last quarter, moves the dial on the second half of that loop. The marketing and the anxiety draw on the same supply. That is not a reason to understate what the systems do, but it does mean an industry that has spent three years arguing its products can do knowledge work should not be surprised that knowledge workers believed it.
The corporate recommendation that follows — explain the role of AI in the strategy early, say plainly how deployment will affect staff — is sound and probably arrives too late to work on its own. The paper's own example shows why. Management adopts AI to grow the business and expects support: the routine tasks go, people move to creative work. Staff already primed toward replacement anxiety read the same memo as cover — the layoffs are being prepared while the workforce is told the technology is for its benefit. Transparency is a lever that requires credibility the anxiety has already spent. A firm that has done the AI announcement before the AI reassurance is working against its own transcript.
The question the study leaves open is whether any of these workers are right. The paper measures perception and says so; what it does not supply, and what the discourse never supplies either, is a way for an individual to check exposure against reality. The calm employees in this account are calm because they think the technology is overblown, not because they have assessed their own task profile. The anxious ones are anxious because they think it works. Both groups are reasoning from headlines. Some of the fear is grounded in fact, some in guesswork from commentators optimising for attention, and nothing in the mechanism distinguishes the two.
Mark Twain's line, which the study's framing invites, is that you cannot depend on your eyes when your imagination is out of focus. A workforce that spends the morning picturing its own redundancy is not looking at the work in front of it. The uncomfortable part for anyone selling this technology inside a company: the employees who correctly understand how fast AI is improving are the same employees least able to concentrate, and persuading staff the systems work is indistinguishable, from where they sit, from telling them they are replaceable.