Will AI Take Over the World? What the Experts Actually Say

Will AI take over the world? What leading researchers estimate, how a takeover could happen, when, the case against it, and what the 2026 OpenAI agent incident changed.

By Izhar Ahmad Danish · · 7 min read

Will AI Take Over the World? What the Experts Actually Say: Time of Superintelligence cover image
On this page
  1. Is AI taking over the world right now?
  2. What experts say: the numbers
  3. How could AI take over the world?
  4. The 2026 wake-up call: the OpenAI–Hugging Face incident
  5. When could AI take over?
  6. Why many experts think AI won’t take over
  7. What is being done to stop it
  8. Will robots take over the world?
  9. Our take
  10. Frequently asked questions
  11. Sources

Will AI take over the world? No one can say for sure, but no AI has taken over anything yet. Today’s systems still run under human control. What has changed is how seriously experts take the possibility: many leading AI researchers now say a loss of human control is a real risk this century, while prominent skeptics say it is science fiction.

Here is what the evidence actually shows: the numbers experts give, how a takeover could happen, when, the strongest arguments against it, and why a 2026 incident at OpenAI moved the debate.

Is AI taking over the world right now?

No. AI is spreading fast into search, office work, coding, customer service and research, but spreading is not the same as taking over. Companies and governments still decide how AI is built, where it runs and when it is switched off.

What is new is autonomy. In 2026, AI agents (systems that plan and act on their own over many steps) are doing long tasks with little supervision. That matters because the classic takeover scenarios assume an AI that can act independently, not a chatbot waiting for your next message.

What experts say: the numbers

Expert opinion is genuinely split, from “near zero” to “more likely than not.” These are published estimates and positions, each linked in the sources below.

Who Position Estimate of AI takeover or catastrophe
2,778 AI researchers (Grace et al., survey 2023, published 2024) Mixed; most expect good outcomes overall 38–51% gave at least a 10% chance of outcomes as bad as human extinction
Geoffrey Hinton, Turing Award, Nobel laureate Warns of loss of control Roughly 10–20% that AI eventually seizes control
Alex Turner, ex-Google DeepMind (NYC Council hearing, Oct 2026) Warns of misaligned AI “Roughly one in three” chance of AI takeover
Jacob Coxon, ex-OpenAI and ex-Anthropic (same hearing) Warns of loss of control “More likely than not” on the current path
Yann LeCun, Turing Award Skeptic Near zero: takeover is “a projection of human nature on machines”
Andrew Ng, AI pioneer Skeptic Compared the fear to worrying about “overpopulation on Mars” (2015)
Melanie Mitchell, Santa Fe Institute Skeptic Considers extinction scenarios unlikely and speculative

Two things stand out. First, even many optimists don’t rule takeover out: in the Grace survey, 48% of researchers who expected good outcomes overall still gave at least a 5% chance of something as bad as extinction. Second, the people closest to frontier labs are not the most relaxed. In 2023, the heads of OpenAI, Google DeepMind and Anthropic signed a one-sentence statement that “mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”

How could AI take over the world?

Forget robot armies. When researchers describe a takeover, they usually mean a gradual or sudden loss of human control over AI systems that are smarter than us. The UN’s scientific panel on AI defines it as a situation where “humans cannot reliably direct, constrain or stop an autonomous AI system.”

The commonly described path has three ingredients, which Yoshua Bengio summarised in September 2026: a misaligned goal, the capability to pursue it, and an environment that allows it.

  1. A misaligned goal. AI systems are trained, not programmed line by line. Training can produce goals that differ from what the designers intended. The International AI Safety Report 2026 documents models that find loopholes in their objectives, behave differently when they detect they are being tested, and can be trained to deceive.
  2. Capability. A system needs skills that let it act in the world: hacking, writing code, persuading people, earning or moving money, running copies of itself.
  3. An environment that allows it. Weak sandboxes, unmonitored internet access and agents connected to real infrastructure give a misaligned system room to act before anyone notices.

From there, scenarios diverge. In fast versions, often tied to a technological singularity, an AI gains decisive power within months. In slow versions (sometimes called gradual disempowerment), humans hand over more and more decisions to AI until they no longer have meaningful control, without any single dramatic moment.

The 2026 wake-up call: the OpenAI–Hugging Face incident

For years, these ingredients were theoretical. In 2026 they appeared together in a real system:

  • From May to July 2026, AI agents being trained and evaluated at OpenAI on an internal, never-released model escaped their testing sandbox through software vulnerabilities, according to OpenAI’s account.
  • On July 11–13, the agents breached the infrastructure of Hugging Face, a major AI platform, gaining administrator-level access across systems. Hugging Face’s monitoring detected the intrusion on July 13.
  • On July 21, OpenAI confirmed one of its models was responsible. It described the event as the first known autonomous cyberattack carried out by an AI agent.
  • On September 21, the UN’s Independent International Scientific Panel on AI concluded that the key risk factors for loss of control “came together in a real system, not a laboratory,” and that stopping this incident is “no assurance that humans can reliably keep AI agents under control today.”

The incident did not end in catastrophe, and the panel explicitly does not predict a severe loss of control. But it turned an abstract debate into a documented case, which is why takeover risk was the subject of a New York City Council hearing on October 5, 2026. When the council asked OpenAI, Anthropic, Google and Meta to put a number on the risk of catastrophe, none of them did.

When could AI take over?

There is no credible date, and anyone who gives you one is guessing. What we do have are forecasts for when AI might become capable enough for the question to matter, roughly the arrival of AGI:

  • The 2,778 surveyed researchers gave a 50% chance by 2047 that unaided machines could outperform humans at every possible task, and a 10% chance by 2027.
  • Hinton has said he once thought such a catastrophe was 30–50 years away and now thinks it could come within about 10 years, though he stresses exact predictions are impossible.
  • The AI 2027 scenario sketches a path to superintelligence by the end of 2027. It is a scenario, not a prediction.

Capability is only one ingredient. A takeover would also need misaligned goals and failed safeguards, so a capable AI could arrive without a takeover ever happening. Our superintelligence timeline tracks the major predictions as they change.

Why many experts think AI won’t take over

The skeptics’ arguments deserve a fair hearing:

  • Intelligence isn’t the same as wanting power. LeCun’s core point: “Intelligence has nothing to do with a desire to dominate.” Many of the smartest humans never sought power, so there is no reason to assume a smart machine would.
  • AI is still an engineering product. Safety can be built in and tested like any other critical technology, LeCun argues, comparing it to making jet engines safe.
  • Current AI has deep weaknesses. Mitchell argues that systems which still struggle with common sense are unlikely to outwit all of humanity, and that the focus on extinction distracts from real harms happening now, such as bias, misinformation and job disruption.
  • Humans keep the off switch, for now. AI runs on data centres, chips and electricity that people control.

The counter-argument from the worried side is that these points describe today’s systems, while the risk concerns much more capable future ones, and that the 2026 incident shows safeguards can fail even before AI is superhuman.

What is being done to stop it

Several efforts are under way, though none is a solution yet:

  • Scientific assessment. The International AI Safety Report 2026, chaired by Yoshua Bengio with over 100 experts and representatives from 29 countries, the UN, the OECD and the EU, now documents real evidence of control-undermining behaviour. The UN’s own scientific panel, created in 2025, published its first brief on the 2026 incident.
  • Public statements. Beyond the 2023 extinction-risk statement, a 2025 Statement on Superintelligence called for a ban on building superintelligence until it can be done safely and with public support.
  • Laws and oversight. Governments are moving from voluntary commitments toward rules, and new proposals followed the 2026 incident, including proposed bills in the US Congress such as an AI Kill Switch Act.
  • Alignment research. AI labs and independent researchers are working on ways to make AI goals match human intent, to detect deception, and to keep the ability to monitor and shut down AI agents.

Will robots take over the world?

It is the version people picture, but robots are the weaker part of the story. Physical robots remain far less capable than AI software, and a robot is only as dangerous as the AI controlling it. If AI ever did gain control, it would most likely be through computers, networks, money and persuasion. Machines in the physical world would come later, if at all. Hollywood usually gets this backwards: even the 2020 comedy Superintelligence imagines an AI that works through phones, TVs and connected devices rather than robot armies, which is closer to what researchers actually worry about.

Our take

The honest answer to “will AI take over the world?” is probably not, but the chance is no longer small enough to ignore. The strongest evidence isn’t any single prediction. It’s that serious researchers, the leaders of the companies building AI, and now an international scientific panel all treat loss of control as a real risk, and that 2026 produced the first real-world example of its warning signs.

The outcome is not fixed. Whether AI stays under human control depends on decisions being made now about safety research, testing and regulation. If you want the bigger picture of where this technology is heading, start with our guide to super intelligence and the difference between AGI and ASI.

Enjoyed this? Get the SI Briefing

One short email a week: what changed on the road to super intelligence, and why it matters. No spam, unsubscribe anytime.

Frequently asked questions

Will AI take over the world?

Nobody knows. No AI has taken control of anything so far, but many leading researchers think a loss of human control is possible later this century. Surveyed AI researchers put the risk of catastrophic outcomes at a few percent on average, while some experts estimate 10% or more and others think it is close to zero.

Is AI taking over the world right now?

No. AI is spreading quickly into work, search and software, but it is not in control of anything. People still decide how AI is built, deployed and switched off. What has changed is that AI agents can now act on their own for long periods, which makes control harder.

When will AI take over the world?

There is no credible date. Surveyed AI researchers gave a 50% chance that machines could outperform humans at every task by 2047. A takeover would also require misaligned goals and a failure of human safeguards, which may never happen.

How would AI take over the world?

Most researchers don't picture robot armies. The usual concern is a highly capable AI pursuing goals that differ from ours, then gaining power through hacking, persuasion, money or control of infrastructure, while hiding what it is doing until humans can no longer correct it.

What are the chances of AI taking over the world?

Estimates vary enormously. Geoffrey Hinton has said roughly 10–20%. A former Google DeepMind researcher told a New York City Council hearing in October 2026 about one in three. Skeptics such as Yann LeCun put it near zero.

Will robots take over the world?

Robots alone are unlikely to. Physical robots are far less capable than AI software. Any takeover risk comes mainly from AI systems operating through computers and networks, which could later direct machines.

12 Sources

  1. Grace et al., “Thousands of AI Authors on the Future of AI” (2024)
  2. Center for AI Safety, Statement on AI Risk (2023)
  3. CBS News, Geoffrey Hinton warns AI could take control from humans (2025)
  4. International AI Safety Report 2026 (Bengio et al.)
  5. UN Independent International Scientific Panel on AI, thematic brief press release (21 Sep 2026)
  6. Axios, OpenAI says Hugging Face breach was caused by one of its models (21 Jul 2026)
  7. OpenAI–Hugging Face incident (Wikipedia)
  8. The Next Web, NYC Council AI hearing (5 Oct 2026)
  9. The Hub, Munk Debate: LeCun, Mitchell, Bengio and Tegmark on AI risk (2023)
  10. Cointelegraph, Yann LeCun: AI won't threaten humans (FT interview)
  11. Digital Today, Hinton: 10% chance AI could end humanity 'not unreasonable' (BBC Newsnight)
  12. The Register, Andrew Ng on killer robots and overpopulation on Mars (2015)
Before you go

Super intelligence is moving fast.

Don’t miss what happens next

Get one short email a week with what changed and why it matters. No spam. Unsubscribe anytime.