NextFin News - Workers inside the artificial-intelligence industry are reporting a distinct form of workplace distress: anxiety, exhaustion and moral strain tied to the fear that the systems they are building pose a genuine threat to society. The toll sits on top of a broader deterioration in tech-worker well-being, where significant burnout rose to 55.7% of respondents in 2026 from 44.7% a year earlier, while career optimism fell from 54.8% to 48.7%.
The distinction matters. Most of the public debate about AI and mental health has focused on job displacement - will the machines take our work? Inside the industry, a second, less discussed stressor is emerging: the psychological cost of building technology whose long-term consequences the builders themselves are not sure are benign. That combination - job anxiety plus societal-risk anxiety - is what makes this moment different from previous technology cycles.
The Situation: A Workforce Building the Future, Worried About What It Is Building
Worker sentiment toward AI has shifted sharply, and the shift is broad-based. In a 2024 survey of workers, a plurality said AI was doing more good than harm across society when it came to finding jobs, building wealth and securing quality of life. By late 2025 that had flipped: 44% of respondents said AI was doing more harm than good, compared with 38% who said it was doing more good. Optimism about AI's impact on workers fell 10 percentage points over the same period, with only 39% now feeling somewhat or very optimistic.
The reversal is not confined to the general public. Pew's survey of 5,273 employed U.S. adults, fielded in October 2024, found 52% worried about the future impact of AI use in the workplace and 32% believing it would lead to fewer job opportunities for them in the long run. The American Psychological Association's 2023 workplace survey found 38% of workers worried that AI might make some or all of their job duties obsolete - and among those worried workers, 64% reported typically feeling tense or stressed during the workday, compared with 38% of those without AI concerns. Nearly two in five workers worried about AI replacement (41%) said they did not matter to their employer, versus 23% of those without such concerns.
The mental-health consequences extend well beyond job anxiety. A 2026 workplace mental-health report based on 1,000 full-time U.S. employees at companies with 250 or more employees, fielded in March 2026, found 48% said their job had negatively impacted their mental health over the past year, 84% reported burnout affecting their productivity at least slightly, and 72% felt pressured to work through mental-health struggles - up 10 points from 2025. Microsoft's 2026 Work Trend Index, surveying 20,000 knowledge workers across 10 markets between February and April 2026, documented the same strain at scale: 81% of leaders reported feeling safe suggesting new ways of working with AI, compared with 67% of employees - a 14-point gap that points to a perception problem at the top of organisations.
Inside the technology sector specifically, the 2026 tech-worker sentiment survey - one of the largest of its kind, capturing thousands of responses across product, engineering, design and research - found burnout climbing 11 points in a single year while 53% of respondents said they would steer a newcomer away from a career in their role. Only 22% cited "losing my job to AI" as their underlying fear; far more worried about being expected to do more for the same pay (51%), getting trapped in an unsustainable pace (46%), and the quality of their work going down (41%). The dominant emotion about AI, the survey found, is ambivalence: 77% of respondents selected at least one positive and one negative emotion.
The Mechanism: Why Building AI Hurts Differently Than Building Other Technology
The core question is why this particular technology generates a different kind of workplace distress. Three transmission channels explain it, and they operate on different time horizons.
First, AI carries an existential-risk dimension that most prior technologies did not. A survey of AI society members projected a 50% likelihood of artificial general intelligence being developed between 2040 and 2065, with 18% of participants believing that AGI development would be existentially catastrophic. The 2023 Expert Survey on Progress in AI, the largest survey of its kind with 2,778 respondents who had published in top-tier AI venues, found that a third to a half of participants assigned 10% or more probability to extremely bad outcomes such as human extinction. For a software engineer, uncertainty about whether your code will be deprecated in two years is one kind of stress. Uncertainty about whether the system you are scaling could cause large-scale societal harm is another - and the second does not resolve when the hiring cycle turns.
Second, the pace of deployment outruns the workforce's ability to adapt, and the adjustment is hitting the youngest workers first. Stanford's Digital Economy Lab, using payroll data, found that employment of workers aged 22 to 25 in the two most AI-exposed occupation quintiles fell about 11% between November 2022 and June 2026, while employment of the same age group in the three least-exposed quintiles grew about 10%. The employment gap for young workers in highly AI-exposed occupations widened to 19% as of June 2026, up from 15% in July 2025, operating primarily through reduced hiring rather than increased separations. That is a compressed adjustment window: the same cohort being asked to build the tools is being told, by the labour-market data, that the entry-level rungs they climbed are disappearing beneath them.
Third, there is a values-dissonance channel. Workers are asked to move fast while the ethical guardrails are still being written. Researchers at the University of Florida have proposed a clinical label for the resulting condition - "artificial intelligence replacement dysfunction," or AIRD - describing symptoms ranging from anxiety, insomnia and paranoia to loss of identity, feelings of worthlessness, resentment and hopelessness among otherwise healthy individuals. The mechanism is not irrational fear; it is the cognitive load of holding two contradictory beliefs at once: that the technology is powerful and valuable, and that it may be dangerous.
AI displacement is an invisible disaster. As with other disasters that affect mental health, effective responses must extend beyond the clinician's office to include community support and collaborative partnerships that foster recovery.
Joseph Thornton, a clinical associate professor of psychiatry at the University of Florida and co-author of the paper proposing the AIRD framework, framed the problem as a public-health challenge rather than an individual one. That framing matters because it points away from resilience training and toward structural responses - upskilling, transparency about how AI will change roles, and channels for workers to raise concerns without retaliation.
The Second-Order Effect: The Mental-Health Cost Becomes a Productivity and Retention Risk
The first-order effect is individual distress. The second-order effect is what that distress does to the industry itself - and this is where the story becomes a market story rather than a wellness story.
Burnout is not a soft metric. When 84% of employees report burnout affecting their productivity and 72% feel pressured to work through mental-health struggles, the output of the sector is being degraded from within. The technology companies racing to deploy AI are simultaneously eroding the psychological capacity of the workforce needed to deploy it safely. This is a self-limiting dynamic: the faster the build-out, the more strain on the builders, and the higher the risk of errors, safety shortcuts and talent flight. A workforce that is ambivalent about its own product does not move at the speed the growth assumptions require.
There is also a selection effect that compounds the problem. The workers most likely to experience this distress are often the ones most thoughtful about the technology's implications - precisely the people the industry most needs in safety and alignment roles. The AI-safety ecosystem is small relative to the scale of the build-out: a 2025 estimate placed roughly 1,100 full-time-equivalent workers at self-identified AI-safety organisations, with a 2026 projection of 2,000 to 2,500 research-fellowship seats across talent programmes. That is a narrow base on which to carry the safety burden of a multi-trillion-dollar deployment, and it is a population reporting elevated mental-health challenges. If the people best equipped to judge the risk are the most likely to burn out or leave, the industry's risk-assessment function degrades exactly as the technology becomes more capable.
The Counter-Thesis: Is This Just Normal Tech-Cycle Anxiety?
The strongest argument against reading too much into this is that every technology cycle produces the same story. The internet, biotechnology, nuclear power and automation have all generated waves of worker anxiety and public fear that later proved overblown. And the tech-worker data is not uniformly bleak: 82% of tech workers say AI is already making them at least moderately better at their jobs, and 60% feel confident or ahead of their peers in AI skills, compared with 22.5% who feel anxious or behind. The majority of the workforce is adapting, not breaking. On this reading, today's distress is cyclical - it will mean-revert as new roles emerge, as workers retrain, and as the technology's benefits become visible in the way they always do.
That counter-argument is partly right but misses the asymmetry between the two stressors. Job-displacement anxiety is cyclical: it mean-reverts as new roles emerge and workers retrain. A worker who fears being laid off can be reassured by a hiring rebound. Existential-risk anxiety is not cyclical in the same way, because it is tied to a belief about the technology's terminal properties, not the labour market's temporary dislocation. A worker who fears the technology itself cannot be reassured by a strong quarter. The two stressors require different remedies, and conflating them leads companies to treat a structural concern with cyclical solutions - resilience training and wellness benefits for a problem rooted in the product's purpose.
The 77% ambivalence figure cuts both ways and is worth sitting with: it means most workers hold the positive and negative views simultaneously. That is not a population on the verge of mass exit; it is a population trying to reconcile genuine benefit with genuine risk. The policy implication is that the distress is real but not necessarily terminal for the industry - it is a friction cost on the build-out, not a strike.
The falsifying signal is specific and observable: if worker sentiment on AI's societal impact reverts to net-positive within the next two survey cycles while deployment continues at the current pace, the structural reading is wrong and this is ordinary cycle anxiety. If, instead, the net-negative reading holds or deepens even as the technology delivers measurable productivity gains, the distress is structural and tied to the technology's perceived risk profile rather than the business cycle. A second signal: if AI-safety hiring keeps pace with deployment headcount, the selection-effect concern weakens; if the gap widens, it strengthens.
What Comes Next: Beneficiaries, the Exposed, and the Signals to Watch
In the short term, the impact falls on employers. Companies that treat AI-related distress as a wellness issue will see it manifest as turnover, absenteeism and safety risk. Those that treat it as a product-governance issue - giving workers channels to raise concerns, tying deployment speed to safety-review capacity, and being transparent about how AI will change roles - will retain the talent they need most. The 14-point gap between leaders and employees on psychological safety around AI suggests plenty of room for management to close the perception divide.
In the medium term, the impact shows up in the pace of deployment. A strained workforce slows down, demands more review, and resists aggressive timelines. That is not necessarily bad for society, but it is a headwind to the growth assumptions priced into the sector. The self-limiting dynamic means the fastest possible build-out is not the most likely one.
In the long term, the impact is on the technology's trajectory. If the builders themselves are a vocal source of caution, political and regulatory pressure for guardrails will come from inside the industry as well as outside it. That changes the shape of the regulatory outcome: it is harder to dismiss safety concerns as Luddism when they come from the engineers who wrote the code.
Short-term, watch the next cycle of the tech-worker sentiment survey and the Microsoft Work Trend Index for whether burnout and net-negative AI sentiment continue to rise. Medium-term, watch AI-safety hiring and retention rates - a widening gap between deployment headcount and safety headcount is the clearest leading indicator of strain. Long-term, watch whether regulatory frameworks begin to cite internal industry warnings; that would mark the point where worker distress becomes a binding constraint on the build-out.
Conclusion
The mental-health toll reported by AI staff is not just a workplace story. It is a signal that the industry's fastest-growing input - human talent - is experiencing the technology's risk profile before the market has priced it in. Job anxiety will fade with the cycle. Fear of what the technology itself might do will not, unless the technology's trajectory changes.
The builders are not just building the future. They are the first people living inside it, and their distress is the earliest measurable data point on what that future feels like.
Explore more exclusive insights at nextfin.ai.
