AI Companies Could Become More Powerful Than Their Host Nations

View the official memo here.

AI companies are on track to produce capabilities that eclipse the power of the United States and other host nations. This is likely to happen in the next few years, and very likely to happen in the next decade.

Leading AI labs explicitly aim to create artificial superintelligence — agentic AI that vastly exceeds human intelligence across all domains — soon. In their words:

Sam Altman (CEO, OpenAI): OpenAI is confident it knows how to build AGI and is “now aiming for superintelligence… in the true sense of the word.”1

Sam Altman (CEO, OpenAI): OpenAI is confident it knows how to build AGI and is “now aiming for superintelligence… in the true sense of the word.”1

Dario Amodei (CEO, Anthropic): powerful AI amounting to “a country full of geniuses in a datacenter,” possibly “as early as 2026.”2

Dario Amodei (CEO, Anthropic): powerful AI amounting to “a country full of geniuses in a datacenter,” possibly “as early as 2026.”2

Demis Hassabis (CEO, Google DeepMind): ~50% chance of AGI within five to ten years.3

Demis Hassabis (CEO, Google DeepMind): ~50% chance of AGI within five to ten years.3

Mark Zuckerberg (CEO, Meta): Meta is “focused on building full general intelligence,” shortly before standing up a ~$14.3B superintelligence effort.4

Mark Zuckerberg (CEO, Meta): Meta is “focused on building full general intelligence,” shortly before standing up a ~$14.3B superintelligence effort.4

Helen Toner (former OpenAI board member), Senate testimony: “What we [in DC] don’t understand is how deadly serious they are about building machines that will outperform humans at everything, and how deadly serious they are that they don’t know if they’ll be able to control the machines that they create.”5

Helen Toner (former OpenAI board member), Senate testimony: “What we [in DC] don’t understand is how deadly serious they are about building machines that will outperform humans at everything, and how deadly serious they are that they don’t know if they’ll be able to control the machines that they create.”5

Dario Amodei: agreed that “the leaders of major A.I. companies are, in fact, likely to be major geopolitical actors,”6 and elsewhere “it feels a little bit wrong for [the whole technology]… to ultimately be in the hands of private actors.”7

Dario Amodei: agreed that “the leaders of major A.I. companies are, in fact, likely to be major geopolitical actors,”6 and elsewhere “it feels a little bit wrong for [the whole technology]… to ultimately be in the hands of private actors.”7

Dario Amodei: “A swarm of millions or billions of fully automated armed drones, locally controlled by powerful A.I. and strategically coordinated across the world by an even more powerful A.I., could be an unbeatable army.”8

Dario Amodei: “A swarm of millions or billions of fully automated armed drones, locally controlled by powerful A.I. and strategically coordinated across the world by an even more powerful A.I., could be an unbeatable army.”8

Sam Altman: called compute “the currency of the future” and expects it may be “the most precious commodity in the world…by the end of this decade.”9

Sam Altman: called compute “the currency of the future” and expects it may be “the most precious commodity in the world…by the end of this decade.”9

Models by OpenAI and Google DeepMind found mathematical insights that have evaded the best human minds for decades.10

Models by OpenAI and Google DeepMind found mathematical insights that have evaded the best human minds for decades.10

Anthropic’s Claude Mythos Preview outperformed a year’s worth of professional bug-finding efforts in a matter of weeks.11

Anthropic’s Claude Mythos Preview outperformed a year’s worth of professional bug-finding efforts in a matter of weeks.11

AIs are able to perform coding tasks in minutes that would take an expert hours, this time horizon is growing exponentially, and evaluators struggle to design new tests.12

AIs are able to perform coding tasks in minutes that would take an expert hours, this time horizon is growing exponentially, and evaluators struggle to design new tests.12

AIs are beginning to hire and manage human workers.13

AIs are beginning to hire and manage human workers.13

RAND: AI could “empower nonexperts to develop weapons of mass destruction”, create strategically decisive “wonder weapons”, or “cause the emergence of artificial entities with their own agency to threaten global security.”14

RAND: AI could “empower nonexperts to develop weapons of mass destruction”, create strategically decisive “wonder weapons”, or “cause the emergence of artificial entities with their own agency to threaten global security.”14

Forethought: AI could enable individuals or small groups to overthrow governments. “An AI-enabled coup is most likely to be staged by leaders of frontier AI projects, heads of state, and military officials; and could occur even in established democracies.”15

Forethought: AI could enable individuals or small groups to overthrow governments. “An AI-enabled coup is most likely to be staged by leaders of frontier AI projects, heads of state, and military officials; and could occur even in established democracies.”15

CAIS: AI-designed superweapons “would upend the nuclear balance entirely and could establish one state’s complete dominance and control, leaving the fate of rivals subject to its will.” Cyber capabilities “could enable individuals to cause disruptions previously achievable only by governments.” Engineered pandemics and bioweapons could “unleash a catastrophe unlike any our biosphere has known.”16

CAIS: AI-designed superweapons “would upend the nuclear balance entirely and could establish one state’s complete dominance and control, leaving the fate of rivals subject to its will.” Cyber capabilities “could enable individuals to cause disruptions previously achievable only by governments.” Engineered pandemics and bioweapons could “unleash a catastrophe unlike any our biosphere has known.”16

Dean Ball & Ben Buchanan (former senior AI policy officials): “A.I. systems have strong and growing capability in materials science, software development and industrial processes — all fundamental to designing and producing all sorts of new weapons,” and “A.I. has become so powerful that, along with its tremendous promise, the technology poses immediate risks to national security.”17

Dean Ball & Ben Buchanan (former senior AI policy officials): “A.I. systems have strong and growing capability in materials science, software development and industrial processes — all fundamental to designing and producing all sorts of new weapons,” and “A.I. has become so powerful that, along with its tremendous promise, the technology poses immediate risks to national security.”17

The race to superhuman AI poses a threat to U.S. interests:

Any actor allowed to develop superintelligence could create capabilities that surpass those of nation-states, including the U.S.

Any actor allowed to develop superintelligence could create capabilities that surpass those of nation-states, including the U.S.

The developers of those capabilities may not retain control of them.18

The developers of those capabilities may not retain control of them.18

A U.S. government project could similarly lead to loss of control.

A U.S. government project could similarly lead to loss of control.

The prudent option is to prevent the development of superintelligence by any actor, as it would upend the current geopolitical regime. This means:

Negotiating an international agreement to halt frontier AI development.19

Negotiating an international agreement to halt frontier AI development.19

Readying countermeasures against the creation of superintelligence worldwide.20

Readying countermeasures against the creation of superintelligence worldwide.20

Developing the authorities and mechanisms to halt domestic actors before loss of control occurs.

Developing the authorities and mechanisms to halt domestic actors before loss of control occurs.

Urgently seeking visibility into the activities and capabilities of frontier AI companies and foreign actors.

Urgently seeking visibility into the activities and capabilities of frontier AI companies and foreign actors.

The U.S. must take steps to prevent foreign powers and private actors from developing superhuman AI, or expect to rapidly lose the ability to defend its interests at home and abroad.

Sam Altman, “Reflections,” Sam Altman (blog), January 5, 2025, blog.samaltman.com, archived at archive.ph/Xpn8T.

Dario Amodei, “Machines of Loving Grace,” October 11, 2024, darioamodei.com, archived at archive.ph/jlYYx.

Demis Hassabis, interview by Steven Levy, “Demis Hassabis On The Future of Work in the Age of AI,” WIRED, June 6, 2025, video, youtube.com.

Mark Zuckerberg, “Just shared Meta’s quarterly earnings report. AI continues to transform everything we do…”, Facebook, April 30, 2025, facebook.com.

Helen Toner, “”Stealth Stealing: China’s Ongoing Theft of U.S. Innovation,” testimony before the U.S. Senate Committee on the Judiciary, April 22, 2026, judiciary.senate.gov.

Dario Amodei and Ross Douthat, interview, “Anthropic’s Chief on A.I.: ‘We Don’t Know if the Models Are Conscious’,” New York Times, February 12, 2026, nytimes.com.

Dario Amodei, interview by Ezra Klein, “What If Dario Amodei Is Right About A.I.?,” The Ezra Klein Show, New York Times, April 12, 2024, podcast, nytimes.com.

Dario Amodei, “The Adolescence of Technology”, January 2026, darioamodei.com.

Sam Altman, interview by Lex Fridman, “Sam Altman: OpenAI, GPT-5, Sora, Board Saga, Elon Musk, Ilya, Power & AGI,” Lex Fridman Podcast #419, March 18, 2024, youtube.com.

Joe Rogero, “The Erdős Proof and AI Capabilities,” Machine Intelligence Research Institute, May 22, 2026, intelligence.org; Google DeepMind, “AlphaEvolve: A Gemini-Powered Coding Agent for Designing Advanced Algorithms,” May 14, 2025, deepmind.google.

Mozilla, “Behind the Scenes: Hardening Firefox,” May 7, 2026, hacks.mozilla.org; Anthropic, “Project Glasswing: An Initial Update,” May 22, 2026, anthropic.com.

METR, “Task-Completion Time Horizons of Frontier AI Models,” accessed June 11, 2026, metr.org.

Andon Labs, “We gave an AI a 3 year retail lease in SF and asked it to make a profit,” April 10, 2026, andonlabs.com; Andon Labs, “Bengt Hires a Human,” February 13, 2026, andonlabs.com; OpenAI, “GPT-4 System Card,” March 2023, openai.com; Jenna Ahart, “AI Agents Are Hiring Human ‘Meatspace Workers’—Including Some Scientists,” Nature, February 13, 2026, nature.com.

Jim Mitre and Joel B. Predd, “Artificial General Intelligence’s Five Hard National Security Problems,” RAND Corporation, February 10, 2025, rand.org.

Tom Davidson, Lukas Finnveden, and Rose Hadshar, “AI-Enabled Coups: How a Small Group Could Use AI to Seize Power,” Forethought, April 15, 2025, forethought.org.

Dan Hendrycks, Eric Schmidt, and Alexandr Wang, “Superintelligence Strategy,” March 5, 2025, nationalsecurity.ai; Casey Aveggio, Aman J. Patel, Sella Nevo, and Kyle Webster, “Exploring the Offense-Defense Balance of Biology,” August 18, 2025, rand.org.

Dean Ball and Ben Buchanan, “This Is What Should Unite the Right and the Left on A.I.,” New York Times, May 4, 2026, nytimes.com.

Geoffrey Hinton, interview by METR, June 27, 2024, youtube.com; Yoshua Bengio, interview by ABC News, July 14, 2023, abc.net.au; Ilya Sutskever and Jan Leike, “Introducing Superalignment,” OpenAI, July 5, 2023, openai.com; Stuart Russell, “DeepSeek, OpenAI, and the Race to Human Extinction,” op-ed, Newsweek, January 31, 2025, newsweek.com.

Aaron Scher, David Abecassis, Peter Barnett, and Brian Abeyta, “An International Agreement to Prevent the Premature Creation of Artificial Superintelligence,” Machine Intelligence Research Institute, November 18, 2025, techgov.intelligence.org.

Dan Hendrycks, Eric Schmidt, and Alexandr Wang, “Superintelligence Strategy,” March 5, 2025, nationalsecurity.ai; Casey Aveggio, Aman J. Patel, Sella Nevo, and Kyle Webster, “Exploring the Offense-Defense Balance of Biology,” August 18, 2025, rand.org.