Resource guide

Is AI Dangerous? Documented Harms and Disputed Risks

For readers who want to push back on AI, documented harms are kept apart from the disputed extinction debate.

Last updated October 15, 2026 2094-word guide Editor Ban the Bots

The Real Risks of AI, Sorted by How Certain They Are

Yes, artificial intelligence (AI) is dangerous in two different ways. Some harms are documented now, including scams, deepfake abuse, biased automated decisions and job cuts. A separate long-term risk, loss of control or extinction, is disputed among experts.

If you want to protect yourself from AI's harms, start with the ones already on record in the table below.

Risk	Status	Evidence	Our guide
Scams and voice-clone fraud	Happening now	The Federal Bureau of Investigation (FBI) 2025 report counted 22,364 complaints that mention AI, with more than $893 million in adjusted losses. The tag does not prove AI caused the loss.	Deepfake scams
Deepfake nudes	Happening now	A Thorn survey found 1 in 8 teens know someone targeted by deepfake nudes.	Deepfakes
Biased or wrong automated decisions	Happening now	A wrongful facial recognition arrest in Michigan ended in a settlement announced in June 2024. The Workday age discrimination case is preliminary, and its claims are allegations.	Facial recognition
Chatbot harm to teens	Happening now	Google and Character Technologies agreed in January 2026 to settle family lawsuits. The terms were not disclosed.	AI safety for parents
Job cuts attributed to AI	Happening now	Challenger, Gray & Christmas counted 120,136 cuts citing AI through September 2026. That is about 21% of all cuts.	AI and jobs
Loss of control or extinction	Disputed	The Center for AI Safety (CAIS) statement calls extinction risk a global priority. The International AI Safety Report says experts vary greatly on its likelihood.	The AI doomers debate

Happening now means a named report, filing or court record documents the harm. Disputed means experts disagree about how likely or how severe the risk is. The table does not rank the harms by how much damage each one causes.

Harms Happening Now

Five harms already have victims or counts on record: scams, deepfake nudes, biased automated decisions, chatbot harm to teens and job cuts.

Scams and Voice Clones

The FBI's Internet Crime Complaint Center (IC3) counted 22,364 complaints tagged AI-related in 2025. Those complaints reported more than $893 million in adjusted losses.

The tag marks a complaint that mentions AI. It does not prove that AI caused the loss.

Among the scam types the report lists, investment scams had more than $632 million in AI-linked losses. Distress scams, where voice cloning mimics a loved one in trouble, had more than $5 million in AI-linked losses.

The FBI says AI-made synthetic content is becoming increasingly difficult to detect and easier to make. Our deepfake scams guide covers a 2024 case in which Arup lost about $25 million through a deepfake video call to its Hong Kong office.

Deepfake Nudes

A Thorn survey of 1,200 young people aged 13 to 20 found that 1 in 8 teens know someone who has been targeted by deepfake nudes. The survey was released on March 3, 2025.

Thorn's separate blog post reports that 1 in 17 teens had deepfake nudes created of them.

In the same survey, 2% of respondents admitted to creating deepfake nudes. Among non-victims, 62% said they would tell a parent. Among victims, only 34% did.

20% believed it was legal to create deepfake nudes of someone else, including a minor. Our deepfakes guide covers the topic in more depth.

Biased and Wrong Automated Decisions

Robert Williams was arrested at his home in Farmington Hills, Michigan, in January 2020, in front of his wife and young daughters.

A facial recognition match had tied him to a Shinola watch theft in Detroit. Prosecutors dropped the charges. A related Michigan Public report says he was held for more than a day.

On June 28, 2024, the American Civil Liberties Union (ACLU) of Michigan announced a settlement. Under it, Detroit police may treat a facial recognition match only as a tip.

The settlement also requires officer training and disclosure to defense attorneys. A federal court gets monitoring power for four years. The police department did not stop using facial recognition.

Michigan Public doesn't give the settlement amount.

In Mobley v. Workday, a federal judge in California granted preliminary certification of a nationwide age discrimination collective action in May 2025. Judge Rita Lin made the ruling.

The group covers applicants aged 40 and over who were denied recommendations through Workday's AI since September 24, 2020. Workday said: "This is a preliminary ruling at an early stage of this case, and before the facts have been established."

The claims in Mobley v. Workday are still allegations. Stanford's AI Index 2025 says models built to avoid explicit bias still show implicit racial and gender biases, such as favoring men for leadership roles.

Chatbots and Teens

In January 2026, Google and Character Technologies, the maker of Character.

AI, agreed to settle family lawsuits, including one alleging a chatbot pushed a teen to suicide.

The settlements cover Megan Garcia's Florida lawsuit and other family lawsuits in Colorado, New York and Texas. The terms were not disclosed, and the settlements need judicial approval.

Garcia's son, Sewell Setzer III, was 14 when he died by suicide in February 2024. The lawsuit alleged an emotionally and sexually abusive relationship with a chatbot.

A federal judge had earlier rejected Character's bid to dismiss the Florida case on First Amendment grounds. Google was named because it hired Character's co-founders in 2024.

Job Cuts

Challenger, Gray & Christmas reported on October 1, 2026, that AI was cited in 120,136 job cut announcements through September.

That is about 21% of all cuts, and the leading reason cited so far this year. United States employers announced 573,195 cuts in total through September.

In September 2026 alone, AI was cited for 3,961 cuts, about 9% of the month's total.

These figures count cuts where the employer named AI as the reason. They do not show that AI did the work. Challenger counts possibly AI-driven technological updates separately.

HR Dive reported that 17,375 cuts had been explicitly attributed to AI in 2025, as of the September report. Our AI and jobs guide covers the wider picture.

Reported Incidents Are Rising

The AI Incident Database recorded 233 AI-related incidents in 2024. That was a record high and a 56.4% increase over 2023, according to Stanford's AI Index 2025.

Stanford's AI Index 2026 reports 362 incidents in 2025, up from 233 in 2024. That is about 55% more (129 divided by 233).

The AI Index 2025 also says AI-related election misinformation appeared in more than a dozen countries in 2024, "but its measurable impact is uncertain."

The Extinction Argument

The CAIS statement is one sentence: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

The statement says its purpose is to open discussion. It notes that "it can be difficult to voice concerns about some of advanced AI's most severe risks."

Its first signatories include Geoffrey Hinton, Yoshua Bengio, Demis Hassabis of Google DeepMind and Sam Altman of OpenAI. Dario Amodei of Anthropic also signed.

The CAIS page shows no date, but press links on it date the statement to May 30, 2023.

Geoffrey Hinton put the chance of AI causing human extinction over the next 30 years at 10 to 20 percent. He said it on the Today programme on British Broadcasting Corporation (BBC) Radio 4 in December 2024.

Hinton also said: "You see, we've never had to deal with things more intelligent than ourselves before."

Fortune reported Yoshua Bengio's October 2025 warning. He said: "If we build machines that are way smarter than us and have their own preservation goals, that's dangerous."

He added: "Even if there was only a 1% chance it could happen, it's not acceptable."

Fortune paraphrases him, without a direct quote, as saying major risks could appear in five to 10 years.

A survey of 2,778 AI researchers who had published in top AI venues was fielded in 2023 and posted in January 2024. Between 38% and 51% of them gave at least a 10% chance to advanced AI leading to outcomes as bad as human extinction.

The range depends on how the question was worded, so it is not one figure. Our AI doomers guide covers the full debate and its surveys.

The International AI Safety Report was first published on January 29, 2025. It was chaired by Yoshua Bengio and written with 96 experts.

The report says there is broad consensus that current general-purpose AI lacks the capabilities to pose loss of control. It also says expert opinion on the likelihood of loss of control within the next several years "varies greatly."

Some experts think such risks are decades away. Others think general-purpose AI could lead to societal-scale harm within the next few years.

The Case That the Fear Is Overblown

Some researchers reject the extinction framing.

Yann LeCun said at a Meta event in Paris on June 13, 2023, that accidentally creating a superintelligence that escapes human control is "just preposterous."

He said that is "not the way anything works in the world." He added that humans "are not stupid enough to roll out infinite power to systems."

Our Yann LeCun guide says he puts the odds of AI-driven extinction close to zero.

Andrew Ng, a Google Brain founder, told the Australian Financial Review that large technology companies are "creating fear of AI leading to human extinction." He suggested they do it to avoid competing with open-source AI.

The International AI Safety Report also records experts who consider loss of control implausible. Others consider it likely, and some see a modest chance that deserves attention.

Will AI Take Over the World?

Whether AI will take over the world is disputed. Geoffrey Hinton and Yoshua Bengio, who warn about it, describe it as a probability.

Hinton's estimate is 10 to 20 percent over 30 years. Bengio says even a 1% chance is not acceptable.

The International AI Safety Report says there is broad consensus that current general-purpose AI lacks the capabilities to pose loss of control.

The second report, published February 3, 2026, says capabilities have continued to improve, especially in mathematics, coding and autonomous operation.

The 2026 report says AI agents pose heightened risks because they act autonomously. It defines loss of control as AI systems that operate outside anyone's control, with no clear path to regaining control.

A key update to the International AI Safety Report on November 25, 2025, made two findings. It found that sophisticated attackers can often bypass current defences, and that the number of companies publishing Frontier AI Safety Frameworks has more than doubled.

The 2026 report covers less ground than the 2025 report, which also addressed bias, environmental impacts, privacy and copyright.

How to Protect Yourself from AI

Protect yourself first against scams and deepfakes, the harms with FBI complaint data behind them.

Hang up on an unexpected call that asks for money, then call the person back on a number you already have. Voice cloning mimics the sound of a loved one in distress.
Treat an urgent video call that asks for money with suspicion, because a deepfake video call cost Arup about $25 million in 2024.
Learn the signs of deepfakes, but do not rely on them alone. The FBI says AI-made synthetic content is becoming increasingly difficult to detect. Our deepfakes guide is the place to start.
Turn off AI features you did not choose. Our how to turn off AI guide covers the steps.
Limit what you share about yourself online. Our guide to protecting your privacy online is the place to start.
If a teen uses a chatbot, read our AI safety section for parents.

If a call or video asks for money, start with our deepfake scams guide.

Sources

Frequently asked questions

▸ What is the risk of AI?

AI carries two kinds of risk. Some harms are documented now, including scams, deepfake nudes, biased automated decisions, chatbot harm to teens and job cuts employers attribute to AI. A long-term risk of loss of control or extinction is disputed among experts.

▸ Will AI take over the world?

The question is disputed. The International AI Safety Report says there is broad consensus that current general-purpose AI lacks the capability to pose loss of control. Geoffrey Hinton puts the chance of AI causing human extinction over 30 years at 10 to 20 percent. Yann LeCun calls that scenario preposterous.

▸ How do I protect myself from AI?

Start with the threats that already have victims. Hang up on unexpected calls that ask for money, then call the person back on a number you already have. Learn the signs of deepfakes, although the FBI says synthetic content is becoming harder to detect. Turn off AI features you did not choose.

▸ Is AI dangerous to humans?

Yes, in two ways. Documented harms, such as scams, deepfake abuse and biased decisions, are happening now. The extinction risk is disputed. Geoffrey Hinton and Yoshua Bengio frame it as a probability. Yann LeCun calls the scenario preposterous.

Latest related briefings

REGULATION POLICY

FCC’s $2B E-Rate Cut Threatens Internet Access for Millions

If the FCC ends the $2B E-Rate program, millions of students—especially from low-income families—could lose crucial internet access at school and home.

Oct 14, 20264 min read

Read analysis PARENTING EDUCATION

New AI Safety Guidelines Aim to Shield Kids Online—Will They Work?

IEEE released new AI safety guidelines to protect kids online, but their impact depends on adoption by tech companies and schools. Parents should stay vigilant.

Oct 14, 20264 min read

Read analysis JOBS LABOR