Advanced AI usually means “frontier” systems: the largest and most capable AI models, such as the chatbots and AI agents built by companies like OpenAI, Google DeepMind and Anthropic. The question is whether the world should stop building ever more powerful versions, for months or even years, until people are confident they can be kept safe.
The idea went mainstream in March 2023, when the Future of Life Institute published an open letter calling on all AI labs to “immediately pause for at least 6 months the training of AI systems more powerful than GPT-4.”1 Elon Musk was one of the signers.25 Days later, a YouGov poll found that 69% of Americans supported a six-month pause on some kinds of AI development.2 No major lab stopped.
The warnings grew louder. In May 2023, Geoffrey Hinton, one of the pioneers of modern AI, left Google so he could speak freely about its risks. “Right now, they’re not more intelligent than us, as far as I can tell. But I think they soon may be,” he said.4 Hinton, Yoshua Bengio and the heads of OpenAI, Google DeepMind and Anthropic also signed a one-sentence statement that begins: “Mitigating the risk of extinction from AI should be a global priority.”3
Experts disagree about how big the danger is. In a survey of 2,778 AI researchers, between 38% and 51% gave at least a 10% chance that advanced AI leads to outcomes as bad as human extinction. Their combined forecast gave human-level machine intelligence a 50% chance of arriving by 2047, 13 years sooner than in the same survey a year earlier.5 The public is more worried than the experts: in 2024, 51% of US adults said they were more concerned than excited about AI, compared with 15% of AI experts.6
Meanwhile, the race has sped up. In January 2025, President Trump announced Stargate, a plan led by OpenAI and SoftBank to spend up to $500 billion on AI infrastructure, a day after he revoked his predecessor's AI safety order.17 “China is a competitor and others are competitors and we want it to be in this country,” he said.17 Less than a week later, a cheaper model from the Chinese startup DeepSeek helped send Nvidia's shares down more than 13% in one morning.13
Governments have mostly chosen rules over pauses. In November 2023, 28 countries including the United States and China, plus the European Union, agreed at Bletchley Park in England that frontier AI has the “potential for serious, even catastrophic, harm, either deliberate or unintentional.”7 The EU's AI Act came into force in August 2024.16 In October 2025, a new statement calling for “a prohibition on the development of superintelligence” drew signers as different as Prince Harry, Steve Bannon and Steve Wozniak.10
The fault lines are sharp. Supporters of a pause say no one should build something they cannot understand or control. Opponents say a pause could not be enforced, would hand the lead to rivals, and would delay breakthroughs like AlphaFold, the AI system behind part of the 2024 Nobel Prize in Chemistry.12
Key terms
Frontier AI
The most capable AI models at any given time, usually built by a small number of well-funded companies.
Artificial general intelligence (AGI)
A hypothetical AI that could learn and do almost any mental task as well as a human can.
Superintelligence
A hypothetical AI that would be far smarter than humans in nearly every area, including science, strategy and persuasion.
AI alignment
The research field that tries to make sure AI systems pursue the goals their designers intend, and not harmful goals of their own.
Existential risk
A risk that could wipe out humanity or permanently destroy its future, like nuclear war or a deadly global pandemic.
Moratorium
An agreed, temporary halt to an activity, often while rules are written or safety is checked.
Precautionary principle
The idea that when an action might cause serious harm, the people who want to go ahead should have to show it is safe first.
Compute governance
Controlling AI by tracking or limiting the powerful computer chips needed to train the largest models.
Pros and cons at a glance
Pro: pause it
Con: keep building
Pro arguments
Con arguments
Key clashes
Most debates about pausing AI are won or lost on these four questions.
1How likely is a disaster?
Nobody can measure the risk of a technology that does not exist yet.
Pro saysMany of the researchers who build AI put a real chance on catastrophe. Even a small chance of extinction is too high to ignore.5
Con saysThese are guesses, not evidence. Most AI experts are more excited than worried, and policy should follow proof of harm.6
2Would everyone actually stop?
A pause that only some follow could be worse than none.
Pro saysScientists paused genetic engineering in the 1970s, and governments already control the chips that AI needs.1126
Con saysRivals like China would keep going, and even signers of the 2023 letter kept building AI.1325
3Pause or regulate?
Both sides want safety. They disagree on whether rules can keep up.
Pro saysLaws move slowly and are easily blocked, as California's vetoed safety bill showed. A pause buys time to write them.15
Con saysRules like the EU's AI Act and licensing of powerful models can manage risk without stopping research.1621
4What does waiting cost?
Every delay has a price. The question is which price is higher.
Pro saysA delay can be made up later. A mistake with a system smarter than us might never be undone.
Con saysAI is already helping science and medicine, as AlphaFold shows. A pause means slower cures.12
Around the world
The US: President Trump revoked his predecessor's AI safety order in January 2025, and in July 2025 the White House released a plan titled “Winning the AI Race: America's AI Action Plan.”1718
The European Union: the AI Act came into force in August 2024, with rules for general-purpose AI models applying from August 2025.16 In 2023, Italy briefly blocked ChatGPT over privacy concerns.24
China: Chinese companies such as DeepSeek compete directly with American labs.13 China also joined the 2023 Bletchley Declaration on AI risks.7
Global cooperation: the International AI Safety Report, chaired by Yoshua Bengio, is backed by more than 30 countries along with the UN, EU and OECD.8
Debate tip: the international picture helps both sides. Pro can argue that only a global agreement makes anyone safe; Con can argue that countries are racing, not pausing, so a pause by one side just changes who leads.
Discussion questions
For class or club discussion. There are no right answers; ask for reasons and evidence.
Should the development of advanced AI be paused? Explain your answer using at least one argument from each side.
How big would the risk of disaster have to be before you supported a pause: 1 in 10, 1 in 100, 1 in a million?
Who should decide how fast AI develops: companies, governments, scientists or voters?
If the United States paused and China did not, would the world be safer or more dangerous? Why?
Is a voluntary pause, like the one biologists agreed to in the 1970s, possible for AI today? What would make it work?
Sources
Future of Life Institute, “Pause Giant AI Experiments: An Open Letter,” March 22, 2023. LinkOpen letter, advocacy group
YouGov, “Would you support or oppose a six-month pause on some kinds of AI development?” daily survey of 20,810 US adults, April 3, 2023. LinkPoll
Center for AI Safety, “Statement on AI Risk,” 2023. LinkPublic statement, research nonprofit
Matt O'Brien, Associated Press report on Geoffrey Hinton leaving Google to warn about AI's dangers, via NY1, May 2, 2023. LinkNews report
Katja Grace and others, “Thousands of AI Authors on the Future of AI,” Journal of Artificial Intelligence Research, vol. 84, 2025 (first released January 2024). LinkPeer-reviewed survey
Pew Research Center, “Views of risks, opportunities and regulation of AI,” from a report comparing the US public and AI experts, April 3, 2025. LinkResearch nonprofit
UK Government, “Countries agree to safe and responsible development of frontier AI in landmark Bletchley Declaration,” November 1, 2023. LinkGovernment statement
UK Government, “First International AI Safety Report to inform discussions at AI Action Summit,” January 29, 2025. LinkGovernment report
Future of Life Institute, “Statement on Superintelligence,” October 2025. LinkPublic statement, advocacy group
Fortune, report on public figures including Prince Harry, Steve Wozniak and Steve Bannon signing a call to ban superintelligence, October 22, 2025. LinkNews report
Nell Greenfieldboyce, “Scientists to meet 50 years after conference that set limits on genetic engineering,” NPR, via Utah Public Radio, February 24, 2025. LinkNews report
“Three scientists win Nobel Prize in chemistry for work on proteins,” Al Jazeera, October 9, 2024. LinkNews report
Maria Aspan, “U.S. stock markets tumble as investors worry about DeepSeek,” NPR, via OPB, January 27, 2025. LinkNews report
IndiaAI, “Andrew Ng and Yann LeCun on why AI research shouldn't be paused,” 2023. LinkGovernment AI portal
Bobby Allyn, “California Gov. Newsom vetoes AI safety bill that divided Silicon Valley,” NPR, via MPR News, September 30, 2024. LinkNews report
European Economic and Social Committee, “General-purpose AI: way forward after the AI Act,” October 29, 2024. LinkEU advisory body
“Trump announces $500bn Stargate venture to build up AI infrastructure,” Al Jazeera, January 22, 2025. LinkNews report
US House Committee on Oversight and Government Reform, “Oversight Committee leaders applaud President Trump's bold plan to cement America's dominance in artificial intelligence,” July 24, 2025. LinkGovernment statement
International Energy Agency, Energy and AI, executive summary, 2025. LinkInternational body
Kristalina Georgieva, “AI Will Transform the Global Economy. Let's Make Sure It Benefits Humanity,” IMF Blog, January 14, 2024. LinkInternational body
Matt O'Brien, “ChatGPT chief says artificial intelligence should be regulated by a US or global agency,” Associated Press, via NBC Bay Area, May 16, 2023. LinkNews report
Anthropic, “Disrupting the first reported AI-orchestrated cyber espionage campaign,” November 2025. LinkCompany threat report
Maxwell Zeff, “OpenAI's o1 model sure tries to deceive humans a lot,” TechCrunch, December 5, 2024. LinkNews report
“ChatGPT available to users in Italy a month after temporary ban,” Al Jazeera, April 28, 2023. LinkNews report
“Elon Musk launches xAI startup to, naturally, understand the universe,” TechRadar, July 12, 2023. LinkNews report
Covington & Burling, “U.S. Imposes Additional Export Controls Restrictions on Advanced Computing and Semiconductor Manufacturing Items,” October 2022. LinkLegal analysis of US rules
Elise Reuter and Jasmine Ye Han, tracker of AI-enabled medical devices authorized by the FDA, MedTech Dive, updated May 27, 2026. LinkTrade news report
Arguments marked “Evidence needed” are waiting for a source before this guide goes live.
