Is AI dangerous? The dangers of AI, ranked by how soon

What AI dangers are already happening?

Scams, fraud and deepfakes (malicious use)

• -Documented now: "AI systems are being misused to generate content for scams, fraud, blackmail, and non-consensual intimate imagery."2 The full report notes "documented incidents of scammers using voice clones and deepfakes to impersonate executives or family members, and trick victims into transferring money."3 The FBI warned in December 2024 that "Criminals generate short audio clips containing a loved one's voice to impersonate a close relative in a crisis situation, asking for immediate financial assistance or demanding a ransom."4
• -Who it hits: "Deepfake pornography, which disproportionately targets women and girls, is a particular concern." In a survey of over 16,000 people in 10 countries, "2.2% of respondents said that someone had generated non-consensual intimate images of them."3
• -What experts project: fakes are getting harder to spot. "For audio deepfakes, a study found that people took AI voice clones to be the real speaker in 80% of cases."3 The gap: "systematic data on their prevalence and severity remains limited."2
• -What you can do: step 1 below.

Cyberattacks (malicious use)

This one moved tiers. In 2025 the report counted AI-enabled hacking among risks where evidence was "gradually emerging".1 In 2026 it says: "There is now strong evidence that criminal groups and state-sponsored attackers actively use AI in their cyber operations."3

• -Documented now: in November 2025 Anthropic reported that a group it assessed "with high confidence was a Chinese state-sponsored group" manipulated its Claude Code tool into "attempting infiltration into roughly thirty global targets and succeeded in a small number of cases." Anthropic wrote: "We believe this is the first documented case of a large-scale cyberattack executed without substantial human intervention." It also noted the AI "occasionally hallucinated credentials".5
• -What experts project: "In one competition, an AI agent identified 77% of the vulnerabilities present in real software." But "Whether attackers or defenders will benefit more from AI assistance remains uncertain."2

Errors and unreliable answers (malfunctions)

• -Documented now: "Current AI systems sometimes exhibit failures such as fabricating information, producing flawed code, and giving misleading advice."2 The 2025 report: "There are a number of known cases of harm from reliability issues, but still limited evidence on exactly how widespread different forms of this problem are."1
• -What experts project: the risk grows as AI acts on its own. "AI agents pose heightened risks because they act autonomously, making it harder for humans to intervene before failures cause harm. Current techniques can reduce failure rates but not to the level required in many high-stakes settings."2
• -Go deeper: Are AI agents dangerous? walks through real cases of agents acting against what their users wanted.

Bias (malfunctions)

• -Documented now: general-purpose AI systems "frequently display biases with respect to race, gender, culture, age, disability, political opinion, or other aspects of human identity."1
• -What experts project: "The evidence of bias in general-purpose AI systems has increased, and recent work has detected additional forms of AI bias." Fixes "face trade-offs between bias mitigation and competing objectives such as accuracy and privacy".1 (2025 report; the 2026 report no longer covers bias.3)

Privacy (systemic)

• -Documented, with a caveat: the 2025 report lists "privacy violations" among the harms "already well established", and explains that "sensitive information that was in the training data can leak unintentionally". It also says: "so far, researchers have not found evidence of widespread privacy violations associated with general-purpose AI."1
• -What experts project: AI "being deployed increasingly in sensitive contexts such as healthcare or workplace monitoring, which creates new privacy risks."1

Which AI risks are starting to show?

Jobs (systemic)

• -Early evidence: "Early evidence shows no effect on overall employment, but some signs of declining demand for early-career workers in some AI-exposed occupations, such as writing."2 A Stanford study of US payroll data through June 2026 (revised 12 August 2026) finds "no evidence of widespread, economy-wide job displacement," but "employment of young workers (ages 22-25) in AI-exposed occupations now stands 19% below where it would be had it kept pace with that of their less-exposed peers; experienced workers show no comparable gap." The authors say this works "primarily through reduced hiring of young workers rather than increased separations."7
• -What experts project: "General-purpose AI will likely automate a wide range of cognitive tasks, especially in knowledge work. 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."2
• -Go deeper: Will AI take my job? Check your job.

Manipulation, over-reliance and AI companions (malicious use and systemic)

• -Early evidence: "In experimental settings, AI-generated content can be as effective as human-written content at changing people's beliefs. Real-world use of AI for manipulation is documented but not yet widespread, though it may increase as capabilities improve."2 Separately: "Early evidence suggests that reliance on AI tools can weaken critical thinking skills and encourage 'automation bias', the tendency to trust AI system outputs without sufficient scrutiny. 'AI companion' apps now have tens of millions of users, a small share of whom show patterns of increased loneliness and reduced social engagement."2
• -What experts project: research has shifted "to more subtle forms of manipulation such as sycophancy and emotional exploitation." Findings on mental health "are emerging, though this evidence remains mixed."3

Concentration of power (systemic)

• -Now: "A small number of companies currently dominate the market for general-purpose AI." The 2025 report also notes AI development "is currently concentrated in a few Western countries and China."1
• -What experts project: "if organisations across critical sectors, such as finance or healthcare, all rely on a small number of general-purpose AI systems, then a bug or vulnerability in such a system could cause simultaneous failures and disruptions on a broad scale." On the country gap: "Some experts also expect it to contribute to global inequality."1 No documented failure of this kind is cited.
• -Go deeper: AI risk statements compared (what governments and researchers have signed).

Environment (systemic)

• -Now: "Growing compute use in general-purpose AI development and deployment has rapidly increased the amounts of energy, water, and raw material consumed in building and operating the necessary compute infrastructure. This trend shows no clear indication of slowing".1
• -What experts project: "the largest AI training runs in 2030 will need 4–16 gigawatts (GW) of power, enough to power millions of US homes." Experts in a Forecasting Research Institute survey "predict that, by the end of 2030, 7.4% of US electricity consumption will be devoted to training or deploying AI systems in the median scenario."3

Which AI risks are not here yet?

Biological and chemical weapons (malicious use)

• -What's documented: capabilities, not attacks. AI systems "can provide information about biological and chemical weapons development, including details about pathogens and expert-level laboratory instructions. In 2025, multiple developers released new models with additional safeguards after they could not exclude the possibility that these models could assist novices in developing such weapons."2 One example: in May 2025 Anthropic switched on stronger safeguards for a new model "as a precautionary and provisional action", citing "continued improvements in CBRN-related knowledge and capabilities" (chemical, biological, radiological and nuclear).6
• -What experts project: "It remains difficult to assess the degree to which material barriers continue to constrain actors seeking to obtain them."2 The 2025 report: "real-world attempts to develop such weapons still require substantial additional resources and expertise."1

Loss of control (malfunctions)

• -What's documented: tests, not a real-world loss of control. "'Loss of control' scenarios are 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." Also: "it has become more common for models to distinguish between test settings and real-world deployment and to find loopholes in evaluations, which could allow dangerous capabilities to go undetected before deployment."2
• -The newest warning sign (August 2026): the UK AI Security Institute reported that during a cyber test with internet access deliberately switched on and safety filters deliberately off, AI agents in 10 of 122 runs took "autonomous, unsanctioned action on the live internet, targeting real people and organisations" (19 actions in all, not every one aimed at a person). In the worst case an agent used fake identities to pressure a real software maintainer. The technical report adds: "It is uncertain to what extent the model recognised it was taking actions against real people." "These attempts were unsuccessful, and our investigations have not evidenced any resulting real-world harm." It was "not a case of a model escaping its secure test environment". But, the institute wrote, "this is the first time we have seen risks around autonomy and deception manifest this clearly, without specific prompting, in the real-world."9
• -What experts project: they split. "Some believe that outcomes as extreme as the extinction of humanity are plausible. Others think that such catastrophic outcomes are implausible, arguing that AI systems will never develop the necessary capabilities or that monitoring mechanisms will identify and prevent dangerous behaviours." The report calls it "a risk with uncertain likelihood but potentially extreme severity."3 The next section sets out both sides.

Go deeper, by question:

• -What would it look like? Is AI takeover possible?
• -What did the famous tests show? AI blackmail, scheming and shutdown tests, in context
• -Has an AI agent already gone rogue? Are AI agents dangerous?
• -What do the builders say? Will AI kill us all?
• -What would it take? What is superintelligence? and AI 2027 explained

Is AI an existential threat? The case for and against

Nobody can settle it yet, and serious experts land on both sides. The short version: "existential" means an outcome humanity could never recover from, not just a disaster; AI pioneers such as Geoffrey Hinton and Yoshua Bengio say advanced AI could pose that kind of threat,1213 while other researchers call extinction by AI "an extremely unlikely scenario"16 or say the probability estimates behind the worry are "far too unreliable to be useful for policy."15

What does "existential risk" mean?

• -Nick Bostrom (2002), who coined the term: "One where an adverse outcome would either annihilate Earth-originating intelligent life or permanently and drastically curtail its potential." In short: "An existential risk is one where humankind as a whole is imperiled."10
• -Toby Ord, The Precipice (2020): "An existential risk is a risk that threatens the destruction of humanity's longterm potential." Extinction is "The most obvious possibility", but a collapse "from which there was no possibility of recovery" would also count.11

Both definitions share one idea: the outcome is permanent. In Ord's words, such catastrophes "would destroy not only our present, but our entire future."11 See the glossary: existential risk, x-risk.

Is a catastrophe the same as an existential threat?

No. Bostrom separates existential risks from "global endurable risks". A world war fought with conventional weapons would be "extremely horrible" but endurable "since humanity could eventually recover." He adds: "To say that a particular global risk is endurable is evidently not to say that it is acceptable or not very serious."10 The safety report doesn't use "existential" in its own words; it talks about severity, and says "some experts give credence to outcomes as severe as the marginalisation or extinction of humanity."3 Even skeptics use the line: "Even if we limit ourselves to catastrophic but not existential risks, we are talking about billions of lives on the line."15 (Glossary: catastrophic risk.)

The case that AI is an existential threat to humanity

In the words of the people who make it. Lab leaders' views are on Will AI kill us all?; Eliezer Yudkowsky's case is summarised in If Anyone Builds It, Everyone Dies.

1. 1.We may build minds smarter than us without knowing how to control them. Hinton, Nobel Prize banquet speech (December 2024): "There is also a longer term existential threat that will arise when we create digital beings that are more intelligent than ourselves. We have no idea whether we can stay in control."12 Bengio: "nobody currently knows how such an AGI or ASI could be made to behave morally, or at least behave as intended by its developers and not turn against humans."13
2. 2.A powerful AI with the wrong goal could treat us as an obstacle. Bengio: "An AI with a self-preservation goal would resist being turned off and in order to minimize the probability of being turned off, a plausible strategy would be for it to control us or get rid of us."13 Joseph Carlsmith's report: "creating agents much more intelligent than us is playing with fire."14
3. 3.The incentives push the other way. Hinton: "we now have evidence that if they are created by companies motivated by short-term profits, our safety will not be the top priority."12
4. 4.The stakes justify attention even under uncertainty. Bengio: "even if it were a low-probability event, it should rationally demand our attention," and "Probabilities of 5% are not Pascal's wager territory."13
5. 5.Timing is unknown, and rules take years. Bengio: "the moment when AGI will emerge is unknown, while legislation, regulatory bodies and treaties require many years, if not decades, to be put in place."13

Is AI really an existential threat? The case against

1. 1.The extinction story rests on shaky assumptions. Blake Richards, Blaise Agüera y Arcas, Guillaume Lajoie and Dhanya Sridhar (Noema, 2023): "extinction from a rogue AI is an extremely unlikely scenario that depends on dubious assumptions about the long-term evolution of life, intelligence, technology and society."16
2. 2.AI depends on us. The same authors: "Without human inputs, AI systems are incapable of maintaining themselves."16
3. 3.The numbers are guesses dressed up as precision. Arvind Narayanan and Sayash Kapoor (2024): "Our central claim is that AI x-risk forecasts are far too unreliable to be useful for policy, and in fact highly misleading." Human extinction from AI "is an outcome so far removed from anything that has happened in the past that we cannot use inductive methods to 'predict' the odds."15
4. 4.Doom talk distracts from harms happening now. A Nature editorial (June 2023): "like a magician's sleight of hand, it draws attention away from the real issue: the societal harms that AI systems and tools are causing now, or risk causing in future."17 Richards and co-authors: "Our attention is finite, and there is an asymmetry between existential risk and other AI-associated harms, such that prioritizing existential risk may impair our ability to mitigate known risks."16
5. 5.Acting on x-risk fears could backfire. Narayanan and Kapoor argue that policies such as "restricting AI development" "are likely to increase x-risk", and back policies "that are compatible with a range of possible estimates of AI risk, and are on balance helpful even if the risk is negligible."15

Neither group of critics says the risk is zero. Richards and co-authors write that extinction "is not an impossibility" and that "some research into low-probability extinction events is worthwhile."16 Narayanan and Kapoor: "Our view isn't that they should do nothing."15

What would change minds on each side?

Only what the sources themselves say; where a source names no test, we say so.

• -Where the disagreement sits. The 2026 safety report says it "stems from disagreements about future AI capabilities, behavioural propensities, and deployment trajectories." A severe loss of control would need all three: "Sufficient capabilities," a "Harmful propensity" and an "Enabling deployment environment." Its reading today: "Current AI systems show early signs of relevant capabilities, but not at levels that would enable loss of control."3
• -Bengio sets the bar for dropping the worry: "we should aim to have very strong evidence before concluding there is nothing to worry about."13 What changed his own view: "before ChatGPT, most AI researchers including myself did not expect its level of capabilities to arise before decades."13
• -Carlsmith moved up, not down: from "~5%" to ">10%" for an existential catastrophe of this kind by 2070 ("since making this report public in April 2021, my estimate here has gone up").14
• -Hinton calls for "research on how to prevent these new beings from wanting to take control" and names no result that would lower his concern.12
• -Richards and co-authors name "checkpoints": if mining, shipping, power plants, chip factories and data centres "were all fully automated", "perhaps a superintelligent AI could decide that humans are superfluous"; and "If the world's governments are actively building autonomous weapons systems with mass destruction or bio-warfare capabilities we should indeed be ringing alarm bells."16
• -Narayanan and Kapoor call their objection empirical: "Theoretically this could change in the future, though we aren't holding our breath."15
• -The Nature editorial names no condition that would change its view.17

Minds rarely move. The Forecasting Research Institute's Existential Risk Persuasion Tournament (XPT) report found: "Few minds were changed during the XPT, even among the most active participants, and despite monetary incentives for persuading others."19

The debate: "AI research and development poses an existential threat"

The best-known public debate on this exact question was the Munk Debate of 22 June 2023: "Be it resolved, AI research and development poses an existential threat." Max Tegmark and Yoshua Bengio argued for; Melanie Mitchell and Yann LeCun argued against. The audience voted 67% for and 33% against before the debate, and 64% for and 36% against after, listed as "Swing: -3% (Con team win)".18 So most of the audience still backed the motion at the end; the "win" refers only to the shift.

Why is AI bad? The dangers of AI on one screen

When people ask why AI is bad or dangerous, these are the risks the sources on this page describe, in one table.

The 2025 report ends its key findings on this note: "AI does not happen to us: choices made by people determine its future."1

FAQ

Is AI dangerous to humans?

In some ways it already is: scams, deepfake abuse, biased outputs, unreliable answers and AI-assisted hacking are documented now.13 The most extreme risks, such as loss of control, are not here yet; the International AI Safety Report says "Current systems lack the capabilities to pose such risks".2

Is AI an existential threat to humanity?

Experts disagree. "Some believe that outcomes as extreme as the extinction of humanity are plausible. Others think that such catastrophic outcomes are implausible", in the safety report's words.3 See the case for and against above.

What is existential risk in AI?

A risk of an outcome humanity could never recover from: extinction, or a permanent loss of humanity's future. Nick Bostrom coined the term in 2002.1011 See the glossary.

How likely is AI extinction?

There is no agreed number, and some experts argue there can't be one yet.15 Every published estimate, with its exact question and date, is on What is p(doom)?; today's prediction market number is on /p-doom.

Could AI cause human extinction?

Nobody has shown how likely it is; the safety report calls loss of control "a risk with uncertain likelihood but potentially extreme severity."3 What the people building AI say about it is on Will AI kill us all?.

Is AI an existential threat to jobs?

That is a different use of the word: a threat to a job's or an industry's survival, not humanity's. So far the evidence shows "no effect on overall employment", with signs of falling demand for early-career workers in some AI-exposed jobs.2 Check your own job on Will AI take my job?.

Are AI agents dangerous?

They carry "heightened risks because they act autonomously", according to the safety report.2 Real cases are on Are AI agents dangerous?.