Fear of AI: Is Artificial Intelligence Really Dangerous?

"Ah, this artificial intelligence... folks, it's really dangerous, you know."

A supplier said that to me a few days ago. No example, no argument behind it: just the feeling. It stuck with me because it's a very common way of talking about the fear of artificial intelligence: vague, taken for granted, as if it were obvious to everyone. The data, as we'll see, says it isn't.

The question that sentence leaves open is simple: dangerous in what way? I set out to line up what surveys say about people's fears, mostly in Italy where I live, and what the data says about how justified those fears are. The result is less reassuring than enthusiasts would like, and very different from what people who fear robots imagine.

Are Italians Afraid of Artificial Intelligence?

Less than you'd think, at least compared with other countries. The Ipsos AI Monitor 2026, published in June with interviews in 32 countries, puts Italy below average on almost every question that measures fear.

Other surveys tell a different story. According to the KPMG and University of Melbourne study from 2025, 84% of Italians are worried about AI's possible negative effects and only 40% are willing to trust it. Pew Research places Italy among the countries with the highest share of people more concerned than excited, around half the population.

The trend, though, is clear. Italian pollster SWG has asked the same question since 2020, and its latest survey, from mid-September, shows positive views of AI falling from 68% to 49%, while negative views rose from 23% to 45%. Only 6% consider it extremely dangerous: most of those who are worried see it as "useful only in some cases, but high-risk."

It isn't just Italy. In the United States, according to Pew Research, 52% of adults said in June they were more concerned than excited, up from 37% in 2021, and for the first time that became a majority among under-30s too.

Each pollster asks different questions, and the numbers don't add up neatly. Above all, surveys measure what people perceive, not how real a risk is: they tell you how much fear there is, not whether it's justified. For that you need other data, which we'll look at fear by fear. First, though, one number that puts the surveys in perspective.

The Italian Paradox: Worried, Without Having Tried It

According to Istat, Italy's national statistics office, 19.9% of Italians aged 16 to 74 used generative AI tools in 2025, against an EU average of 32.7%. Italy is second to last in Europe, ahead of only Romania.

The interesting part is why. Among those who never used it, 59.9% said they didn't need it and 21.6% said they didn't know how. Only 5.5% held back because of security or data protection concerns.

So in Italy the fear is mostly a feeling: it comes from the news, headlines, small talk. It rarely comes from direct experience. That's the same tone as my supplier's remark, and it's why it's worth taking the fears apart one by one.

Fear Number One: Will Artificial Intelligence Take Our Jobs?

It's the most widespread fear and the easiest to understand. The ninth Eudaimon-Censis report, presented at the end of July, found that 42.6% of Italian workers fear AI will take people's places. In the United States, according to Gallup, 79% expect fewer jobs over the next ten years, and in Pew's survey 71% expect fewer over the next twenty.

This is where it pays to separate two things the press mixes up constantly: what has already been measured and what is forecast.

The Challenger figure needs care: these are the reasons companies write in press releases, and "AI" sounds better in a press release than "we over-hired." The Stanford figure is measured on payroll data, and it's the most serious one. It says the first impact falls on the young person looking for a first job, not on the worker with twenty years of experience. It also shows the decline is concentrated where AI replaces tasks, while employment holds or grows where AI complements them. And the gap widened over the past year.

The Anthropic Economic Index from January adds a useful detail: on Claude.ai, 52% of uses support the person doing the work and 45% automate the task. When businesses use it through the API, automation rises to about 75%. In other words, individuals tend to get help, while companies tend to replace whole steps.

The fear about jobs is justified, but not in the way it's usually told. The concrete risk today isn't mass unemployment: it's fewer entry points for people starting out, and tasks disappearing inside jobs that remain.

It's Not the First Time

In 1930 John Maynard Keynes wrote about "technological unemployment" as "a temporary phase of maladjustment." Economist James Bessen studied ATMs in the United States: between 1985 and 2002 they went from about 60,000 to more than 350,000, and over the same period bank tellers didn't decline, they increased slightly, because opening a branch became cheaper and their work shifted toward advising customers.

Every historical parallel has a limit, though: past technologies mostly affected manual, repetitive work, while this one reaches the work of people who write, analyze and decide. And it moves much faster.

The Most Concrete Danger: Scams and Deepfakes

Of all the fears, this is the one with the most evidence behind it: documented cases, police reports, money lost.

In February 2025 someone cloned the voice of Italy's defense minister, Guido Crosetto, and called some of the country's best-known business leaders asking for money to free kidnapped journalists. Former Inter Milan owner Massimo Moratti paid about €980,000, later frozen in a Dutch account, as Sky TG24 reported. A year earlier an employee of the engineering firm Arup in Hong Kong had transferred the equivalent of $25 million after a video call in which every "colleague" was a deepfake.

You don't need to be a famous businessperson. According to Italian postal police data reported by Milano Finanza, between January and August 2025 there were about 12,000 online scams, with €143 million stolen. Fake online trading accounted for 29% of cases but 81% of the money lost, and the ads used to lure victims featured fake videos of the Italian prime minister.

Europol puts it plainly in its 2025 organized crime report: AI doesn't invent new crimes, it makes existing ones more effective.

Three Defenses That Work Even If You Know Nothing About Tech

A family password. If a voice that sounds like a child or a parent urgently asks for money, ask for the agreed word. A cloned voice won't know it.

Always call back. Hang up and dial the number saved in your contacts, not the one that called you.

Distrust famous faces promising returns. No politician, journalist or famous entrepreneur advertises investment platforms on social media. If you see one, it's a fake video.

Misinformation and Elections: The Risk Is Confusion

According to the KPMG study, 50% of Italians fear AI-manipulated elections and 67% no longer know whether to trust what they find online. Analyses of the 2024 elections by the Centre for Emerging Technology and Security at the Alan Turing Institute found no evidence that AI changed any outcome. They did find something else: content that reinforces existing beliefs and feeds confusion and hate online.

So the harm is subtler than a manipulated vote. It's the habit of believing nothing anymore, real videos included.

Privacy Risks: The Fear That Worries Italians Least

This is where Italians are calmer than average: 61% trust companies to protect their data, against 47% across the 32 Ipsos countries. Yet it's also where Italian authorities have stepped in the most. The data protection authority fined OpenAI €15 million in December 2024, blocked DeepSeek in January 2025 and fined the Replika app €5 million in 2025.

Here the fear is misdirected more than exaggerated: the problem lies less in AI "spying on us" and more in what we tell it ourselves, from work documents pasted into a chat to health data.

"Will It Make Us Dumber?"

There's also a fear that mostly concerns people who already use AI. Two studies get cited constantly. The first is MIT's "Your Brain on ChatGPT": people who wrote essays with ChatGPT showed less brain activity and remembered less of what they'd written. It's a study of 54 people and hasn't been peer-reviewed yet. The second, by Microsoft and Carnegie Mellon, presented at CHI 2025 with 319 knowledge workers, found that the more people trust AI, the less they exercise critical thinking.

I touched on this when writing about AI hallucinations and blind trust in AI answers: the risk lies in handing over the checking too. Young people share the same fear. In a survey by INDIRE, Italy's national education research institute, presented in September and covering 5,342 students, 94% use AI, 84% for homework too, and 44% fear it will reduce their ability to think independently. More than half, 53%, have used it at least once to confide in it or vent.

The studies are small or rely on what people report about themselves, but they all point the same way: the risk grows when AI replaces reasoning instead of supporting it. The Microsoft study ties the drop in critical thinking to trust in the tool, not to using it as such.

Is This the End of Humanity?

This is the fear that makes the most headlines. Geoffrey Hinton, 2024 Nobel laureate in physics and one of the fathers of neural networks, told the BBC, as reported by The Guardian, that there's a 10 to 20% chance AI leads to human extinction within thirty years. In June 2025 Yoshua Bengio founded LawZero, a nonprofit working on safer systems, pointing to signs of deception and self-preservation in the most advanced models.

On the other side, Yann LeCun has called these scenarios "complete B.S.," and two Princeton researchers, Arvind Narayanan and Sayash Kapoor, argue that AI should be treated as a "normal technology": powerful, but bound to spread slowly, like electricity.

Experts disagree on this, and there's no measurable evidence today that settles it either way. What can be measured is something else: the International AI Safety Report, signed by more than a hundred experts and updated in February 2026, lists fraud and cyberattacks among the risks already here. It's the same reason Anthropic decided in April not to release its most capable model publicly, as I wrote about in Project Glasswing: it was too good at finding flaws in software. These are risks that affect people today, not in thirty years.

The Risks and Dangers of Artificial Intelligence: How Real Each Fear Is

The Rules Are Coming, Slowly

Part of the fear comes from the feeling that nobody is in control. That's not entirely true. The EU AI Act has been in force since August 1, 2024: the most serious bans have applied since February 2025, and the rules for large general-purpose models since August 2025. In Italy, Law 132 of 2025 created a specific crime for spreading deepfakes that cause harm, punishable by one to five years in prison.

But rules move slower than technology: at the end of July the EU regulation known as the "Digital Omnibus" postponed to 2027 and 2028 the obligations for high-risk systems, some of which were due to kick in on August 2. And Italians want rules: in the SWG survey, 46% want development to continue with stricter controls, and only 5% think the race shouldn't slow down.

Fear of AI: Frequently Asked Questions

Why is artificial intelligence scary?

Because it changes fast and people mostly hear about it through headlines and films. Istat data shows that only 19.9% of Italians aged 16 to 74 have used it: for most, it's a technology known by hearsay. And according to SWG, opinions got worse precisely in the years it was talked about most: positive views fell from 68% in 2020 to 49% in 2026.

Is artificial intelligence dangerous for humans?

In some ways, yes, and those ways are documented: scams with fake voices and videos, fewer hires for young people in the most exposed jobs, risks to personal data. Other widely cited dangers, such as human extinction, divide experts and have no measurable evidence today.

What are the most concrete risks of AI today?

Scams using deepfakes and cloned voices are the risk with the most evidence: cases like the fake Crosetto call and, according to the Italian postal police, €116 million stolen through fake online trading between January and August 2025 alone. Next come the effects on young people's jobs and the loss of reasoning skills when people lean on the tool too much.

Will AI take our jobs?

So far, measured data doesn't show mass unemployment. It shows a narrower effect: according to Stanford, employment among 22-to-25-year-olds in the most exposed jobs is 19% lower than among their peers, while experienced workers haven't been affected.

What are the risks of AI in schools?

According to the INDIRE survey, 94% of Italian students use AI and 44% fear it will reduce their ability to think independently. The available studies link the risk to use that replaces reasoning, not to use as such.

How can you protect yourself from AI scams?

With three habits: an agreed family password for urgent requests for money, always calling back the number saved in your contacts, and distrusting any famous face on social media promising easy returns.

A Fair Fear, but Without Arguments

With the evidence lined up, my supplier's remark isn't wrong: it's unfounded. The data says artificial intelligence is dangerous mainly when it reaches a phone with the voice of a relative or the face of a politician, not when it shows up in movies. And that, for now, it weighs more on people looking for their first job than on those who already have one.

Vague fears have one flaw: they don't tell you what to do. Specific ones do. A family password, a number to call back, a bit of suspicion toward videos that look too perfect, and the habit of checking what AI tells us are worth more than any debate about extinction.

So in the conversations around you, when someone says artificial intelligence is dangerous, what reason do they give? Do they give one at all?