AI dread: a reasonable warning, or a fear that holds us back?

AI dread: a reasonable warning, or a fear that holds us back?

You can use AI to finish something faster and still feel uneasy about what that means for your future.

Perhaps the tool helps with a report you dislike writing. Then you wonder whether your employer sees the saved time as room for better work or a reason to need fewer people. Perhaps you enjoy generating images but worry about what happens to creative careers. Those positions can coexist.

“AI dread” is a useful name for that discomfort. It describes more than frustration with another software update: the feeling that work, trust, skills and control are changing before we have agreed on the terms.

I think we should take the concerns seriously without treating dread as a reliable forecast. The strongest response combines scrutiny of real harms with room to discover real benefits. Neither compulsory enthusiasm nor permanent alarm helps us make good decisions.

What does “AI dread” mean?

For this article, AI dread means anticipatory unease about how artificial intelligence could affect your livelihood, independence, relationships or sense of purpose. It is an informal umbrella term here, not a diagnosis or a single established scientific measure.

Related research uses “AI anxiety”. A 2025 German validation study of the Artificial Intelligence Anxiety Scale distinguishes concerns about learning AI, job replacement, humans being sidelined, and humanoid AI. Its authors also note missing cut-offs for identifying high anxiety and uncertainty about which specific technologies respondents have in mind. A survey score should not be turned into a diagnosis.

It helps to separate neighbouring ideas:

• Anxiety involves worry about what might happen.
• Fatigue is exhaustion with the volume of tools, announcements and expectations.
• Scepticism is doubt about claims or evidence. It can be entirely calm.
• Opposition can be a considered ethical or political position, rather than fear.
• Existential-risk concern addresses potential severe future threats. It is one part of the debate, not a synonym for every worry about AI.

I found the phrase in commentary and marketing, but no defensible evidence identifying a single originator. In July 2026, The San Francisco Standard reported that Notion was behind “AI Dread?” billboards and a hotline. That shows the phrase being used publicly; it does not establish who coined it or how many people experience it.

What do people actually think?

There is no single public verdict, and using AI does not necessarily mean trusting its wider consequences.

In Pew Research Center’s April 2025 report, 51% of US adults were more concerned than excited about AI’s increasing use in daily life, while 11% were more excited. Among the AI experts surveyed, 47% were more excited and 15% more concerned. In both groups, 38% were equally concerned and excited.

The experts were not uniformly relaxed. Seventy per cent were extremely or very concerned about people getting inaccurate information from AI, and 55% were highly concerned about bias, the same proportion as the public on bias.

The gap matters. People building or studying a technology may anticipate benefits that others cannot yet see. They may also have different exposure to its costs and rewards. These survey results cannot tell us which explanation accounts for the gap.

Gallup’s 2026 US Gen Z survey offers another useful distinction. Among 1,572 respondents aged 14–29, surveyed from 24 February to 4 March, 51% reported using generative AI at least weekly. Yet excitement had fallen to 22% and anger had risen to 31% compared with the previous year. Anxiety remained steady at 42%.

That last detail deserves care. “AI anxiety is rising” would misrepresent this particular finding. Anger rose; anxiety did not. Even daily users had become less excited and hopeful.

Frequent users were generally more positive than nonusers, but that association does not prove that using AI cures anxiety. People who already feel comfortable with it may be more likely to use it.

For a UK perspective, the Ada Lovelace Institute’s June 2025 Licence to build briefing synthesises public-sector research involving nearly 16,000 people in surveys and around 400 in deeper qualitative studies. Its central finding is conditional support: people want evidence, explanations, involvement and ways to challenge decisions. This is a synthesis across studies, not a single poll of “AI dread”.

These sources describe different populations and questions. None supplies a worldwide prevalence figure for the term.

Why does the dread exist?

Jobs and the distribution of benefits

A claim that AI will increase productivity does not answer a worker’s question about income or security. Higher output could support growth, better service, reduced hours or redundancies. Decisions by employers and policymakers influence which outcome people experience.

The International Labour Organization’s 2025 analysis estimates that about one in four jobs worldwide has some exposure to generative AI. It stresses that exposure means potential changes to tasks, not an equivalent number of job losses. Transformation is the more likely overall pattern in its assessment.

That is a reason to reject simplistic replacement headlines. It is not a guarantee that nobody will lose work. A job can survive while its pay, autonomy, entry requirements or daily character change substantially.

Skills, identity and the pressure to keep up

People invest years in becoming good at something. A tool that reproduces part of the visible output can make that investment feel less secure, even when the tool cannot assume the whole role.

There is also a practical learning burden. “Everyone should use AI” leaves unanswered which tools are approved, when to train, who checks mistakes, and what happens if the tool fails. Constantly changing expectations can make preparation feel impossible.

An essay titled The texture of AI dread interprets the unease through meaning, agency and human connection, rather than money alone. That is a philosophical interpretation, not proof of a universal psychological mechanism. It still raises a useful question: would assurances about income settle every concern about losing valued work? I doubt they would.

Unreliable systems making consequential decisions

Some worries concern errors people can already encounter, rather than imagined future intelligence.

In a December 2023 statement about the FTC’s case against Rite Aid, Commissioner Alvaro Bedoya described allegations that facial-recognition systems produced thousands of incorrect matches, leading to customers being wrongly searched, accused or expelled.

That case concerns facial recognition, not a generative chatbot. The distinction matters. The shared concern is institutional reliance on a system that can be wrong, particularly when an affected person has little practical ability to contest it.

Conflicting promises and little control

A person can reasonably support an AI-assisted translation tool and oppose automated decisions about access to essential services. The level of choice, consequence and recourse differs.

The Ada Lovelace Institute’s findings suggest that trust depends partly on the institutions deploying AI and whether people can influence its use. Telling people to become more technologically confident cannot substitute for answering who is accountable.

Public messaging complicates matters further. Notion’s co-founder Akshay Kothari told The Standard that he sees AI as augmenting people and removing busywork. The same report describes sceptical reactions to an AI company sponsoring a campaign about AI anxiety. Both are understandable: recognising fear can open a conversation, but reassurance from a seller is not independent evidence that the risks are resolved.

The case for taking AI dread seriously

There are benefits to listening to concern. That does not mean distress itself is desirable.

It can reveal requirements a product team missed. A customer asking “Can I reach a person?” may be identifying a necessary escalation route. An employee refusing to upload client records may have noticed a data-handling risk. Calling either reaction resistance could hide useful information.

It challenges promises about who benefits. Productivity measures tell us something about output. They do not automatically tell us whether gains become higher wages, improved service or concentrated profits. Asking about distribution is a legitimate economic question.

It can protect skills and meaningful choice. Teams need to decide which activities people should still practise, which decisions need human ownership, and where an alternative to AI should remain available. Those choices deserve discussion before habits and dependencies become difficult to reverse.

It puts responsibility on institutions. Employers can provide paid training, realistic expectations, worker consultation and honest communication. Product companies can provide evidence, appeal routes and limits on consequential actions. These are more useful responses than requiring everyone to demonstrate enthusiasm.

The case against letting dread set the agenda

The strongest counterargument is that broad fear can become too imprecise to guide action.

It can turn possibility into certainty. “Some tasks are exposed” becomes “my profession will disappear”, with the intermediate assumptions omitted. Capability, reliability, adoption, cost, regulation and customer acceptance all affect outcomes. Serious long-term risks deserve assessment; an uncertain scenario should not be presented as a settled deadline.

It can obscure demonstrated benefits. In the November 2024 version of Generative AI at Work, Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied a staggered rollout involving 5,172 customer-support agents. AI access increased issues resolved per hour by 15% on average, with larger benefits for less-experienced and lower-skilled workers. The researchers also found evidence of improved worker learning and customer interactions.

That is evidence from a particular workplace, not a promise for every job. The most skilled workers saw small declines in conversation quality. Still, blanket rejection would overlook a setting where assistance produced measurable benefits.

It can encourage avoidance or compulsive monitoring. Refusing every low-risk experiment leaves some practical questions unanswered. Following every prediction can consume attention without improving a decision. These are plausible traps, not outcomes established for every anxious person by the surveys above.

It can pathologise disagreement. The label “AI dread” becomes harmful if it allows an organisation to treat concerns about consent, copyright, surveillance or employment as an emotional problem to manage. A calm refusal can be well reasoned. Better messaging does not resolve a bad policy.

It can flatten very different technologies into one verdict. A voluntary drafting assistant, a fraud detector and an automated hiring system need different evaluations. General optimism makes this mistake too.

Should we care? Yes, without demanding that people feel differently

My view is that concern should change what we ask and test. It should not, by itself, determine whether every AI application is accepted or rejected.

For product managers, this means researching perceived risk alongside usefulness. Ask users to describe the specific outcome they fear. Is it wrong information, exposure of private data, reduced independence, job insecurity or being unable to speak to a human? Those answers imply different product and organisational changes.

Consider a proposed customer-support assistant. A sensible pilot would let staff review suggestions, forbid unsupported promises, provide escalation, and measure resolution quality alongside handling time. It would also examine review burden and staff experience. Management should explain how the results will affect staffing rather than allowing a demo to stand in for that conversation.

Success would not be “employees stopped worrying”. It would be better service under acceptable working conditions, with evidence that the safeguards work. A decision to redesign or stop the pilot should remain possible.

For an individual, I would use a short decision note:

1. Name the specific concern and the decision it affects.
2. Separate what has happened from what someone predicts.
3. Identify what evidence or protection would change your assessment.
4. Choose a proportionate next action: a bounded experiment, a training request, an objection, a conversation with management, or no adoption yet.
5. Set a time to revisit the decision rather than monitoring announcements continuously.

No one can individually solve a labour-market transition through better prompting. Personal learning can help; fair employment practices and public accountability still matter. If worry is persistently affecting sleep or daily functioning, discussing it with a qualified professional is reasonable, regardless of which technology prompted it.

The next time a colleague says they feel uneasy about AI, ask what they are afraid will happen. Then decide whether the appropriate response is evidence, a product safeguard, an organisational commitment or stopping the proposed use. Reassurance should follow those answers.

---

Research note: Sources reviewed on 14 September 2026. This article combines public-opinion research, workplace research, a UK policy synthesis, regulatory allegations, reporting and explicitly labelled commentary. Survey findings describe their stated populations; forecasts and occupational exposure are not observed job losses. The pros, cons and practical recommendations are the author’s synthesis. “AI dread” is used descriptively, not diagnostically.