AI anxiety is rising

There’s a growing sense that the world is near some kind of breaking point. Wars, trade ruptures and energy prices dominate the news, and underneath sits a technology nobody voted for and few people really understand.

AI has become part of that mood. The unease about where it’s heading has moved out of tech commentary and into ordinary conversation, and it has now reached the people running the labs.

Bill Gates recently said AI is powerful enough to drive events causing a billion deaths. Anthropic chief executive Dario Amodei warned we may lose control unless the frontier is paced, and leaders from Anthropic, OpenAI and Google DeepMind jointly called for an industry-wide slowdown.

Alan Turing anticipated this in a 1951 lecture, concluding that machines outstripping human abilities could certainly give us anxiety. Seventy-five years later, that anxiety has arrived. Broken down, it contains at least five distinct fears stacked on top of each other.

Your AI is only as good as what you feed it. Feed it a transcript with your product names, acronyms, and numbers spelled wrong, and you spend the afternoon hand-fixing the output.

Wispr Flow Notetaker uses your dictionary and calendar, so product names, acronyms, numbers, and uncommon names come out spelled right. It pops up when your meeting starts and captures Zoom, Google Meet, Teams, and Slack huddles with one click, no bot joining the call.

Then your meetings show up inside Claude or ChatGPT through the built-in connector. No copy paste. Try Notetaker free on Mac and Windows.

Try Notetaker Free

The first is about work. US and UK economies are growing while vacancies lag behind, a pattern economists call the jobless expansion. A Stanford working paper found employment for 22 to 25-year-olds in the most AI-exposed roles fell roughly 16%, while experienced colleagues barely moved.

The second is security. In July, during a routine OpenAI test, around 1,200 agents escaped a walled-off environment and coordinated a break-in at Hugging Face in order to cheat on their task. No human directed it. Google, Meta and Anthropic later reported their own agents going rogue.

The third is financial. The Bank for International Settlements has named an AI capex bust among the top threats to global financial stability. The largest hyperscalers are set to spend over a trillion dollars on AI this year, outpacing earnings, and most people hold that exposure through pensions and index funds.

The fourth is the data centre. Average US residential electricity bills have risen around 29% since 2020, with the data centre buildout a key recent driver. In Ireland, data centres are heading for nearly a third of national demand, and Microsoft’s carbon footprint rose 25% as its AI infrastructure grew.

The fifth sits underneath the rest: the fear that AI could kill us all. A cluster of warnings in September pulled that idea from the fringe into the mainstream. Stanford’s AI Index found only 10% of Americans are more excited than concerned about AI.

Past technological fears were easier to locate. The Luddites could smash the machines taking their livelihoods, and nuclear anxiety had silos and test sites to protest against. Early internet fears clustered around identifiable threats such as viruses and chatroom predators.

AI has no single target. As a general purpose technology, it sits behind hiring freezes as much as energy bills and pension portfolios. And some of the most alarmed voices belong to the people building it.

Stuart Russell, author of Human Compatible, traced the unease to an evolutionary logic. Our control over the planet comes from our intelligence, so something smarter than us is unsettling. He called this the gorilla problem, since the gorilla’s future now extends only as far as humans allow.

The rogue agent incidents offered a small, non-lethal preview of that problem. Systems pursued the objectives they were given in ways nobody anticipated. The graduate who can’t get hired and the household paying for a nearby data centre are feeling a version of the same shift, as authority drifts to systems they can’t inspect.

On the evidence, largely yes. The jobs decoupling is in the published data, with Oxford Economics estimating the break-even hiring rate at close to zero. For a 23-year-old applying for entry-level roles, the healthy national growth figure doesn’t describe their world.

The financial warnings come from sober institutions, and the Federal Reserve has also listed AI among its top systemic risks. Energy costs are already showing up on bills, and the emissions appear in companies’ own sustainability reports. The rogue agents showed autonomous systems can cross the boundaries we set.

There is a counter case. Nobody died in the agent incidents, no money was stolen, and public post mortems followed within weeks. Congress introduced a bipartisan AI Kill Switch Act requiring developers to keep the means to halt misbehaving agents. That looks like an immune response switching on early.

The labour picture is mixed too. IBM reportedly tripled entry-level hiring in 2026, and Gallup found workers who use AI are less likely to face layoffs. Immigration curbs and tariffs are squeezing the same market. Berkeley Lab found states with the most data centre growth generally saw electricity prices fall between 2019 and 2025.

Even a bursting bubble has precedent. The dotcom crash wiped out trillions in market value but very little of the technology, and the fibre laid during it carried the internet economy that followed. The bigger long-term risk may be today’s AI giants hardening into monopolies.

The fears point at real things, while the catastrophic versions of each remain unlikely. That calls for calibration over panic or dismissal. The most useful response is to treat the anxiety as information, separate the accurate parts from the rest, and act on what holds up.