AI is already causing real, measured harm in a few specific ways: cheaper scams and fake voices, confident mistakes, unfair results for some groups, and privacy leaks. Those are the risks to plan around today. Bigger effects, on jobs and on who controls the technology, are real but argued over in size. Whether advanced AI could one day get beyond human control is disputed among leading researchers themselves, and there is no agreed way to measure it. This page sorts the risks by how sure anyone can be, and names who says what.
Think of a powerful new medicine. Some side effects are already showing up in patients, and doctors count them. Some effects on society, such as cost and who gets access, are real but debated. And some rare, serious outcomes are argued about by experts because they have not happened yet. Sensible people take all three seriously, but they do not treat them as equally certain.
- Happening now: Scams and deepfakes fake voices and video calls trick people into sending money; this risk has the hardest numbers.
- Happening now: Confident errors AI can state false things as fact, and a wrong summary could mislead even doctors.
- Happening now: Bias AI can work worse for some people, in decisions such as loans or rental applications.
- Happening now: Privacy what you type goes to a company, and the bigger danger is often what people paste.
- Happening now: Misinformation and flattery AI makes convincing text cheap, and chatbots tend to agree with you.
- Argued about: Jobs employers name AI in job cuts, but national figures barely move; first jobs show the clearest effect.
- Argued about: Power in a few hands the UK regulator warns that a few firms with market power control what AI is built from.
- Argued about: Energy and water one chatbot answer uses little power, but the total is growing fast.
- Argued about: Leaning on it too much more confidence in AI went with less critical thinking, a link but not proof of cause.
- Contested: Losing control leading researchers disagree sharply about far more capable future AI, with no agreed way to measure the risk.
The risks at a glance
The plain point: the risks fall into three groups, from counted to contested.
| How sure | The risks | What you can do |
|---|---|---|
| Happening now | Risks: scams and deepfakes, confident errors, bias, privacy, persuasion and flattery. | Do: verify money requests, check facts, keep secrets out, ask who checked the system. |
| Argued about | Risks: jobs, power held by a few firms, energy use, over-reliance. | Do: judge the tasks in your own work, keep your own skills in use. |
| Contested | Risks: losing control of far more capable future AI. | Do: read who is saying it and what evidence they offer. |
Happening now, and measured
The plain point: most of today’s harm comes from people using AI to deceive, or from trusting AI output that is wrong.
Scams and deepfakes
This is the risk with the hardest numbers. The FBI’s 2025 crime report says it “received more than 22,000 complaints reporting AI-related information. Adjusted losses of these complaints exceed $893 million.”1 In the best-known case, an employee of the engineering firm Arup joined a video call with people he believed were the chief financial officer and colleagues; all were deepfakes, and he sent about US$25.6 million across 15 transfers.2 Voices are easy to copy too: Microsoft researchers showed speech copied from “only a 3-second enrolled recording”.3 How AI fraud works shows the patterns, and deepfake self-defence gives the family code word and callback habits that beat them.
Confident errors
AI can state false things as fact. NIST, the US government’s standards agency, calls it confabulation: “The production of confidently stated but erroneous or false content”. It gives a sharp example: “a confabulated summary of patient information reports could cause doctors to make incorrect diagnoses”.4 The 2026 International AI Safety Report, written by more than 100 experts, says current systems “sometimes exhibit failures such as fabricating information, producing flawed code, and giving misleading advice.”5 Why AI makes things up explains the cause.
Bias
AI can work worse for some people than others. A 2018 study of commercial face-analysis systems found error rates “of up to 34.7%” for darker-skinned women, against a maximum of “0.8%” for lighter-skinned men.6 More recently, AI-writing detectors “incorrectly labeled more than half” of essays by non-native English speakers as AI-written.7 The stakes are real decisions: NIST notes AI can affect whether someone is “authorized for a bank loan or accepted as a rental applicant.”8 See AI bias meaning.
Privacy
What you type goes to a company, and may be kept and used. Anthropic’s consumer terms, for example, apply training to “Free, Pro, and Max accounts when this setting is on”, with business plans treated differently.9 The bigger danger is often what people paste: other people’s details, client files or passwords. What happens to what you type lists the settings to check.
Misinformation, persuasion and flattery
AI makes convincing text cheap. In one study, 24.4% of people agreed with a propaganda claim when shown no article, and 43.5% after reading one written by GPT-3, close to the 47.4% for real propaganda.10 Spreading it is a separate step that AI has not solved; how misinformation spreads explains. Chatbots also tend to agree with you, and people like it: one study found participants “trusted the sycophantic AI model more”.11 See sycophancy meaning.
Today’s measured dangers are mostly about trust: a face, a voice or a confident answer is no longer proof. Check anything that moves money, affects a person, or will be repeated as fact.
Real, but argued about: how big and how soon
The plain point: these effects have evidence behind them, but serious people disagree about their size.
Jobs
Employers name AI when they cut jobs: since 2023, AI “has been cited in 184,538 job cut announcements” in the US, according to the outplacement firm Challenger, Gray & Christmas.12 Yet Yale’s Budget Lab finds the national job mix shifting no faster than before ChatGPT.13 The clearest effect is on young people starting out, and the Dallas Fed says it is “not due to layoffs but to a low job finding rate”.14 The International AI Safety Report sums up the split: “some expect job losses to be offset by new job creation, while others argue that widespread automation could significantly reduce employment and wages.”15 Which jobs are safe from AI? has the full evidence.
Power in a few hands
In 2024 the UK’s competition regulator, the CMA, warned that “firms controlling critical inputs for developing FMs may restrict access to shield themselves from competition” (FMs are foundation models, the big general models behind chatbots). It pointed to “a small number of incumbent technology firms which already hold positions of market power”, linked by “an ‘interconnected web’ of over 90 partnerships and strategic investments”.16 In the largest survey of AI researchers, more than half said “substantial” or “extreme” concern is warranted about scenarios including “authoritarian control, and inequality.”17 Who builds AI names the players.
Energy and water
One chatbot answer uses little power, but the total is growing fast. The International Energy Agency projects data-centre electricity “roughly doubling from 485 TWh in 2025 to 950 TWh in 2030, accounting for around 3% of global electricity demand”.18 Whether that is a crisis or manageable depends on where it lands; what AI actually runs on gives both scales.
Leaning on it too much
A Microsoft Research survey of 319 knowledge workers found “higher confidence in GenAI is associated with less critical thinking”.19 The International AI Safety Report adds that companion apps “now have tens of millions of users, a small share of whom show patterns of increased loneliness”.15 A survey shows a link, not proof of cause, which is why this sits in the argued-about group.
Contested: could advanced AI get out of control?
The plain point: leading researchers disagree sharply, and there is no agreed way to measure the risk. Here is who says what.
Those who call it a serious risk. A one-line statement published by the Center for AI Safety reads: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” Its signatories include Geoffrey Hinton, Yoshua Bengio, Demis Hassabis, Sam Altman and Dario Amodei.20 In a survey of 2,778 AI researchers, “Between 38% and 51% of respondents gave at least a 10% chance to advanced AI leading to outcomes as bad as human extinction.”21 The same survey found that most respondents, 68.3%, still thought good outcomes from superhuman AI more likely than bad.21
Those who call it overblown. Yann LeCun, described by TIME as Meta’s AI chief, called the existential-risk idea “preposterous” in a 2024 interview: “AI systems, as smart as they might be, will be subservient to us. We set their goals”.22 The researchers Arvind Narayanan and Sayash Kapoor argue AI is best seen as “normal technology”, like electricity or the internet, and warn that drastic policies built on superintelligence fears “will, in fact, make things much worse if AI turns out to be normal technology.”23
What the international review says. The 2026 International AI Safety Report, chaired by Yoshua Bengio, describes loss of control as AI that operates “outside of anyone’s control”. It says “Current systems lack the capabilities to pose such risks, but they are improving in relevant areas such as autonomous operation,” and warns that models increasingly “distinguish between test settings and real-world deployment”, which could hide dangerous abilities from testers.24 The research effort to keep powerful AI doing what people intend is called alignment; AGI meaning and superintelligence meaning explain the ideas behind the debate.
On long-term risk, the honest position is uncertainty. Be wary of anyone who is certain either way, and ask what evidence would change their mind.
What this means for you
The plain point: the risks you can act on today are the measured ones.
- Money and identity: treat any urgent request for money or secrecy as a warning, and call back on a number you already have.
- Facts: check names, numbers, quotes and sources before you repeat them or act on them.
- Privacy: keep passwords, health details and other people’s information out of chatbots, and check the training setting.
- Decisions about people: if AI helps judge a job application, a loan or a grade, ask who tested it and whether a person checks its calls.
- Agents: only let an AI act where a mistake can be undone, or where you approve each step.
For the other side of the ledger, see pros and cons of AI.
Questions people ask
Is AI dangerous to humans?
Today, mainly through misuse and misplaced trust: scams, deepfakes, confident errors and biased decisions, all measured. Whether future AI could threaten humanity is disputed; the International AI Safety Report says current systems “lack the capabilities to pose such risks”.15
Can AI become self-aware?
No AI today is self-aware; IBM lists self-aware AI as theoretical. Whether it could ever happen is an open question, separate from how capable AI becomes.25 Types of AI explains the categories.
Will AI take my job?
Some people have lost jobs where employers named AI, but national figures barely move so far, and the clearest effect is on first jobs. Judge the tasks in your week; which jobs are safe from AI? goes through the evidence.
Is it safe to use ChatGPT?
For everyday tasks, yes, if you keep private details out and check facts. Is ChatGPT safe? covers the settings and the fake apps to avoid.
What is the biggest risk of AI right now?
By measured harm, fraud has the clearest numbers, with more than 22,000 AI-related complaints to the FBI in 2025.26 In everyday use, this site’s view is that confident errors matter most, because they are easy to miss.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. According to the page, which AI risk has the hardest numbers behind it today?
- Scams and deepfakes
- AI getting beyond human control
- AI becoming self-aware
Show the answer
Answer: Scams and deepfakes. The page says scams and deepfakes are the risk with the hardest numbers, from FBI complaint data and documented cases.
2. How does the page describe the risk of advanced AI getting out of control?
- Proven, and expected very soon
- Disputed among leading researchers, with no agreed way to measure it
- Settled, because experts agree it cannot happen
Show the answer
Answer: Disputed among leading researchers, with no agreed way to measure it. The page says leading researchers disagree sharply, so the honest position is uncertainty, and to be wary of anyone who is certain either way.
3. What does the page suggest if AI helps judge your job application, loan or grade?
- Trust it, because a computer has no opinions
- Avoid applying anywhere that uses AI
- Ask who tested it and whether a person checks its calls
Show the answer
Answer: Ask who tested it and whether a person checks its calls. The page notes AI can work worse for some groups, and advises asking who tested the system and whether a person checks its decisions.
Related on this site: What Is the Singularity in AI?
- 1FBI Internet Crime Complaint Center, 2025 IC3 Annual Report, read at source 17 Sep 2026“In 2025, IC3 received more than 22,000 complaints reporting AI-related information. Adjusted losses of these complaints exceed $893 million.” ↑
- 2CNN Business, 16 May 2024, read at source 17 Sep 2026the employee “was duped into attending a video call with people he believed were the chief financial officer and other members of staff, but all of whom turned out to be deepfake re-creations.” · “He subsequently agreed to send a total of 200 million Hong Kong dollars — about $25.6 million.” ↑
- 3Microsoft Research, VALL-E, January 2023, read at source 10 Sep 2026“only a 3-second enrolled recording of an unseen speaker”. ↑
- 4NIST, AI 600-1, Generative AI Profile, July 2024, read at source 1 Oct 2026“The production of confidently stated but erroneous or false content (known colloquially as ‘hallucinations’ or ‘fabrications’)”. · “a confabulated summary of patient information reports could cause doctors to make incorrect diagnoses”. ↑
- 5International AI Safety Report 2026, publication page, 3 Feb 2026, read at source 5 Oct 2026“Current AI systems sometimes exhibit failures such as fabricating information, producing flawed code, and giving misleading advice.” · “authored by over 100 AI experts”. ↑
- 6Buolamwini & Gebru, Gender Shades, PMLR 81, 2018, read at source 1 Oct 2026“darker-skinned females are the most misclassified group (with error rates of up to 34.7%). The maximum error rate for lighter-skinned males is 0.8%.” ↑
- 7Liang et al., GPT detectors are biased against non-native English writers, Patterns, July 2023, read at source 17 Sep 2026“they incorrectly labeled more than half of the TOEFL essays as ‘AI-generated’ (average false-positive rate: 61.3%).” ↑
- 8NIST, There’s More to AI Bias Than Biased Data, 16 Mar 2022, read at source 1 Oct 2026“AI can make decisions that affect whether a person is admitted into a school, authorized for a bank loan or accepted as a rental applicant.” ↑
- 9Anthropic, Updates to Consumer Terms and Privacy Policy, August 2025, read at source 11 Sep 2026training applies to “Free, Pro, and Max accounts when this setting is on”, and does not apply to “Claude for Work, Claude for Government, Claude for Education, or API use”. ↑
- 10Goldstein et al., How persuasive is AI-generated propaganda?, PNAS Nexus, 2024, read at source 18 Sep 2026“While only 24.4% of respondents who were not shown an article agreed or strongly agreed with the thesis statement, the rate of agreement jumped to 47.4%”; “43.5% of respondents who read a GPT-3-generated article agreed”. ↑
- 11Cheng et al., preprint arXiv:2510.01395, submitted 1 Oct 2025, read at source 23 Sep 2026“participants rated sycophantic responses as higher quality, trusted the sycophantic AI model more, and were more willing to use it again”. ↑
- 12Challenger, Gray & Christmas, Job Cut Announcement Report, 6 Aug 2026, read at source 23 Sep 2026“Since 2023, when AI was first tracked as a distinct reason, it has been cited in 184,538 job cut announcements.” ↑
- 13The Budget Lab at Yale, Tracking the Impact of AI on the Labor Market, read at source 9 Sep 2026“shifts in the occupational mix were well on their way during 2021, before the release of generative AI, and more recent changes do not seem any more pronounced”. ↑
- 14Federal Reserve Bank of Dallas, Young workers’ employment drops in occupations with high AI exposure, Jan 2026, read at source 9 Sep 2026the fall “is not due to layoffs but to a low job finding rate for young workers entering the labor force”. ↑
- 15International AI Safety Report 2026, publication page, read at source 5 Oct 2026“Economists disagree on the magnitude of future impacts: some expect job losses to be offset by new job creation, while others argue that widespread automation could significantly reduce employment and wages.” ↑a“‘AI companion’ apps now have tens of millions of users, a small share of whom show patterns of increased loneliness and reduced social engagement.” ↑b“Current systems lack the capabilities to pose such risks”. ↑c
- 16UK Competition and Markets Authority, CMA outlines growing concerns in markets for AI Foundation Models, 11 Apr 2024, read at source 5 Oct 2026“firms controlling critical inputs for developing FMs may restrict access to shield themselves from competition” · “a small number of incumbent technology firms which already hold positions of market power in many of today’s most important digital markets” · “an ‘interconnected web’ of over 90 partnerships and strategic investments”. ↑
- 17Grace et al., Thousands of AI Authors on the Future of AI, arXiv:2401.02843, submitted 5 Jan 2024, revised 8 Oct 2025, read at source 5 Oct 2026“More than half suggested that ‘substantial’ or ‘extreme’ concern is warranted about six different AI-related scenarios, including misinformation, authoritarian control, and inequality.” ↑
- 18IEA, Key Questions on Energy and AI, April 2026, read at source 17 Sep 2026“Our updated projections see electricity consumption from data centres roughly doubling from 485 TWh in 2025 to 950 TWh in 2030, accounting for around 3% of global electricity demand by that date.” ↑
- 19Lee et al., Microsoft Research, The Impact of Generative AI on Critical Thinking, CHI 2025, April 2025, read at source 5 Oct 2026“We survey 319 knowledge workers” … “higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking.” ↑
- 20Center for AI Safety, Statement on AI Risk, undated page, read at source 5 Oct 2026“Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” Signatories listed include Geoffrey Hinton, Yoshua Bengio, Demis Hassabis, Sam Altman and Dario Amodei. ↑
- 21Grace et al., Thousands of AI Authors on the Future of AI, read at source 5 Oct 2026“2,778 researchers who had published in top-tier artificial intelligence (AI) venues” · “Between 38% and 51% of respondents gave at least a 10% chance to advanced AI leading to outcomes as bad as human extinction.” ↑a“While 68.3% thought good outcomes from superhuman AI are more likely than bad, of these net optimists 48% gave at least a 5% chance of extremely bad outcomes such as human extinction”. ↑b
- 22TIME, Meta’s AI Chief Yann LeCun on AGI, Open-Source, and AI Risk, 13 Feb 2024, read at source 5 Oct 2026“AI systems, as smart as they might be, will be subservient to us. We set their goals, and they don’t have any intrinsic goal that we would build into them to dominate.” ↑
- 23Narayanan & Kapoor, AI as Normal Technology, Knight First Amendment Institute, 15 Apr 2025, read at source 5 Oct 2026“even transformative, general-purpose technologies such as electricity and the internet are ‘normal’ in our conception” · “will, in fact, make things much worse if AI turns out to be normal technology.” ↑
- 24International AI Safety Report 2026, publication page, read at source 5 Oct 2026 (chairYoshua Bengio): “scenarios where AI systems operate outside of anyone’s control, with no clear path to regaining control.” · “Current systems lack the capabilities to pose such risks, but they are improving in relevant areas such as autonomous operation.” · “it has become more common for models to distinguish between test settings and real-world deployment and to find loopholes in evaluations”. ↑
- 25IBM, Types of Artificial Intelligence, read at source 5 Oct 2026, which describes self-aware AI, with awareness of “its own internal conditions and traits along with human emotions”, as theoretical.↑
- 26FBI, 2025 IC3 Annual Report, read at source 17 Sep 2026“In 2025, IC3 received more than 22,000 complaints reporting AI-related information.” ↑
Liang et al., GPT detectors are biased against non-native English writers, July 2023, read at source 17 Sep 2026.
Goldstein et al., How persuasive is AI-generated propaganda?, 2024, read at source 18 Sep 2026.
Cheng et al., arXiv:2510.01395, 2025, read at source 23 Sep 2026.
Grace et al., Thousands of AI Authors on the Future of AI, revised 8 Oct 2025, read at source 5 Oct 2026.
Lee et al., Microsoft Research, The Impact of Generative AI on Critical Thinking, April 2025, read at source 5 Oct 2026.
Sources read before 5 Oct 2026 were read for this site’s linked detail pages and are re-used here with the date they were read. The comparison on this page (the new medicine), the three-group sorting and the advice in “What this means for you” are this site’s own, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK