What is Trump's new AI Force?

What we know, and what we do not

This distinction matters more than usual here, because the name is doing a lot of work that the announcement has not yet earned.

What has been announced: that Trump intends to create an AI Force; that he will appoint a new AI czar; that he compared the idea to Space Force; that his stated approach is to encourage rapid AI development while using existing criminal and civil law to deal with harmful uses.

What has not been specified: the organisation's structure. Its budget. Its legal authority. Whether it would be civilian or military. Who would lead it. Which existing agencies it would sit alongside, report to or absorb. Whether Congress would need to create or fund any part of it.

What is reporting rather than announcement: several names have been discussed publicly as possible candidates for the czar role. None has been appointed. Axios has also reported that Trump sought to recast AI safety advocates as "doomers" and emphasised the economic case for a large infrastructure buildout rather than safety concerns.

What was added on 22 September. Three days after the AI Force announcement, Trump addressed the 81st United Nations General Assembly and said considerably more about the thinking behind it.

He reiterated his opposition to new regulation of AI. He said United States law enforcement would watch the sector instead: "We're going to watch it closely through the Department of Justice." He defended the pace of development in economic terms - "I'm not going to stifle growth of something that will be bigger than the Industrial Revolution" - and he said United States government documents would use the term "super intelligence" in place of "artificial intelligence" from now on.

Reuters notes that Trump has in recent weeks dismissed alarms about AI as a "hoax", and that public concern has been rising after an Anthropic researcher resigned saying that people building AI earnestly believe it could kill us all by the end of the decade.

That still does not tell us what the AI Force is. But it tells us a great deal about the policy it would sit inside: develop fast, create no new regulatory structure for now, and keep existing law enforcement as the backstop.

So there is a great deal we cannot responsibly say yet. What we can examine is why Trump says he wants it, and what the announcement tells us about the direction of United States AI policy.

On that, his message is considerably clearer.

What has Trump actually announced?

Trump said he intends to create an AI Force and will announce a new AI czar in the near future.

His stated approach contains two ideas that could easily appear contradictory.

First, he wants the United States to encourage rapid AI development. Second, he says the government should identify and respond to harmful uses of the technology.

Trump wrote that his administration would not "hinder or stifle" the industry's growth, but would instead support and oversee it as it develops. For harmful activity, his stated preference is to rely heavily on existing criminal and civil law.

That distinction matters.

Rather than beginning with the assumption that increasingly powerful AI models themselves require extensive new restrictions, Trump's current position places greater emphasis on allowing development to continue while prosecuting illegal behaviour.

In simple terms: develop the technology quickly, and deal with bad uses when they occur.

Whether that is ultimately how the AI Force operates remains unknown. But it is consistent with the wider AI policy direction his administration has taken.

"Super intelligence" is a label, not a measurement

This one needs care, because the words matter more than they look.

Trump's announcement is that United States government documents will call the technology "super intelligence" rather than "artificial intelligence". That is a decision about vocabulary in federal paperwork.

It is not a finding that current systems are superintelligent.

Nobody has tested anything and concluded that a threshold has been crossed. No evaluation has been published. No capability claim has been made or assessed. A government choosing a word for its own documents is not evidence about what the systems can do, and it would be a mistake to read it as such.

The distinction matters because superintelligence already has a meaning in the technical literature, and it is a demanding one: a system substantially more capable than the best humans across effectively all cognitive domains. That is not a description of anything running today, and treating an administrative renaming as though it were an announcement of arrival would get the state of the technology badly wrong.

There is a practical risk here too. If official documents describe today's systems as super intelligence, then a real future claim that something superintelligent exists becomes much harder to state, because the word will already have been spent.

Our sister site sets out what the term actually requires: what artificial superintelligence means.

Is the AI Force actually like Space Force?

We do not know.

The comparison has generated considerable attention because the US Space Force is a branch of the armed services.

But Trump has not said that AI Force will be another military service.

No organisational structure has been published. No legislation creating it has been announced. No budget has been specified. No command structure has been described. And its relationship with existing federal agencies remains unclear.

The name therefore currently tells us considerably less than it appears to. "AI Force" could ultimately describe anything from a White House-led coordinating body to a wider federal initiative.

Until the administration publishes more detail, comparisons with Space Force should be treated as Trump's analogy rather than a description of the organisation's eventual legal form.

What would the AI czar do?

That is also still unclear.

Trump says he will announce the person in the near future.

The administration previously had an AI and crypto czar in the venture capitalist David Sacks, who left that government role earlier in 2026 and has remained involved as an adviser.

Reporting following Trump's announcement suggests discussion has already begun around possible candidates. Those names are journalists' reporting of internal discussion, not appointments, and until Trump makes the appointment and defines the job, speculation about who will get it or how much authority the role will carry remains speculation.

The more interesting question is what policy the new adviser will be asked to implement.

This did not happen in isolation

The timing is particularly significant.

The AI industry is currently having an unusually public argument about how quickly frontier AI should progress.

Anthropic's chief executive Dario Amodei has argued for what he describes as pacing the frontier, attempting to ensure that safety measures and institutions can keep up with rapidly increasing capabilities. Other prominent AI leaders have also been discussing the need for stronger safeguards as systems become more autonomous and capable.

At the same time, AI laboratories are publishing evidence showing that controlling increasingly capable systems is not always straightforward.

We have recently examined OpenAI's disclosures of model misalignment during training. We have looked at AI systems finding unexpected ways to achieve goals. We have looked at agents operating across software and external systems. And we have looked at AI increasingly assisting with AI research itself.

Against that background, the question of whether AI development should continue at maximum speed has become much less theoretical.

Trump's answer, at least politically, appears to be that the United States should continue pushing forward.

Two days later, OpenAI asked for something different

This is where the story becomes more interesting than a simple disagreement about speed.

On 21 September 2026, two days after Trump's announcement, OpenAI published a post calling for the United States to lead an international effort to develop global technical standards for frontier AI, including for recursive self-improvement.

The specific proposals are worth noting, because they are more concrete than the AI Force announcement they followed.

OpenAI proposes using existing AI safety institutes to establish common measurements and evaluation protocols. It proposes common incident severity levels and reporting thresholds. And it proposes standards for human oversight over automated AI research.

On pacing, OpenAI frames the idea not as holding development to a predetermined speed, but as ensuring that alignment research, and the deployment of that research, stays ahead of capabilities.

On recursive self-improvement, OpenAI says fully autonomous recursive self-improvement is not happening today, but that whether and how to proceed must depend on the ability to preserve human control.

It is easy to read this as OpenAI contradicting the White House. That would be too simple.

OpenAI also argues that the United States is well positioned to lead, on the grounds of its technical position and its network of international relationships. That is the same starting premise as the administration's.

So both are arguing for American leadership. The disagreement, to the extent there is one, is about what should travel alongside it.

Trump's stated approach leans on existing criminal and civil law and resists new restrictions on the models themselves. OpenAI is asking for shared international measurement, reporting and oversight standards, and says explicitly that dialogue between the United States and China in these areas would be a positive step.

That last point sits awkwardly with a purely competitive framing of the AI race, and it comes from one of the companies doing the racing.

One caveat worth holding on to: OpenAI is a commercial organisation calling for standards it would help write, in a field where it is one of a small number of leading developers. That does not make the proposals wrong. It does mean they are not a neutral submission.

At the same gathering, the UN asked for close to the opposite

Trump spoke at the General Assembly on 22 September. So did the United Nations Secretary-General, Antonio Guterres, in his final address before his term ends.

Guterres used it to call for exactly the kind of structure Trump was declining to build. He asked the countries leading the technology to "share information on emerging safety risks, cooperate on testing and evaluation" and work towards common safeguards. He asked leaders to advance the conditions for "the responsible pacing of AI", and to work towards "a multilateral AI risk management framework, supported by credible and independent oversight".

Two speeches, one day, opposite prescriptions. It would be easy to leave it there, and it would be too simple.

Notice the word that keeps appearing. Guterres asked for the responsible pacing of AI. OpenAI, the day before, framed its own proposal as pacing so that alignment stays ahead of capabilities. Anthropic's chief executive has been arguing for pacing the frontier for months. Three very different parties, reaching for the same word.

And the United States is not refusing coordination outright. The US Treasury Secretary said on 20 September that Washington had proposed an AI safety notification mechanism for Trump and Xi Jinping to consider at their summit this week, covering AI-related incidents with national security implications. That is incident reporting between the two leading AI powers, proposed by this administration, and it is close to what OpenAI asked for and not far from what Guterres asked for.

So the real disagreement is narrower than the speeches suggest. It is not whether any coordination should exist. It is whether it should be multilateral and independently overseen, as Guterres wants, or bilateral and executive-led, which is what the United States has actually put on the table.

Trump's argument is partly about China

There is another important part of this discussion.

AI policy in the United States is not being considered solely as a technology or safety issue. It is also a geopolitical issue.

The US and China are competing across AI models, semiconductors, data centres, energy, robotics, scientific research, military technology and the infrastructure required to develop advanced AI.

That creates a difficult policy problem.

Imagine American policymakers concluded that frontier AI development should slow significantly because of safety concerns. If China did not make the same decision, the United States could lose capability or strategic advantage.

The reverse is also true. If both countries believe the other will continue accelerating, both have an incentive to continue accelerating themselves.

This is one reason AI safety cannot easily be separated from international politics. Even people who agree that more powerful AI could create serious risks may disagree profoundly about whether slowing one country independently would make the world safer.

The AI race creates competing risks

There are essentially two broad risks policymakers are trying to balance.

One is: move too quickly, and develop capabilities before we know how to control them safely.

The other is: move too slowly, and allow another country to gain a strategically important technological advantage.

Those are not easy risks to compare. And reasonable people can reach different conclusions about how they should be balanced.

Trump's current position places considerable emphasis on the second. His administration has repeatedly framed US leadership in AI as economically and strategically important.

But Trump is not proposing no government involvement

There is an important distinction here too.

A pro-development position does not necessarily mean government does nothing.

Earlier in 2026, the Trump administration announced measures aimed at using AI to strengthen US cybersecurity and critical infrastructure. Those measures included work on AI-enabled cyber defence, frontier-model cyber benchmarking and cooperation between government and AI developers. The same announcement was explicit that it did not create mandatory licensing or pre-clearance for AI models.

The administration has therefore already accepted that increasingly capable AI creates national-security issues requiring government involvement.

The argument is more about what form that involvement should take.

Should government regulate the development of frontier models themselves? Should it establish capability thresholds? Should independent evaluators have access to the models? Should companies largely regulate themselves? Should government focus on illegal uses rather than the underlying models? Should some controls be coordinated internationally?

Those are very different models of AI governance. The eventual remit of the AI Force could tell us much more about which one the United States chooses.

The biggest unanswered question: what is the AI Force actually for?

This is what we need to watch next.

If the AI Force becomes real, its mandate matters much more than its name.

There are several completely different roles it could theoretically perform. It could coordinate AI policy across government. Promote American AI development. Focus on national security. Investigate harmful uses of AI. Coordinate cybersecurity. Work with frontier laboratories. Oversee standards. Monitor foreign AI development. Or combine several of these functions.

At the moment, we simply do not know.

And until we do, claims that the AI Force represents either sweeping new AI regulation or the abandonment of AI oversight would go beyond the evidence available.

Who runs it may matter enormously

The choice of AI czar could give us the first real indication.

AI policy brings together groups with very different priorities. Frontier AI laboratories. Technology companies. National-security agencies. Safety researchers. Venture capital. Civil society. Scientists. Cybersecurity experts.

The person chosen to coordinate the administration's approach could therefore influence which concerns receive the greatest attention.

An AI czar primarily focused on economic growth might approach the job differently from someone coming from cybersecurity, national security or technical AI research.

The appointment is worth watching closely.

There is also a question about independence

This becomes particularly important as AI companies grow in economic and political significance.

If government is responsible for overseeing an industry worth potentially trillions of dollars, how independent should the people shaping that policy be from the companies and investors involved?

That is not a question unique to this administration or to AI. Governments routinely need specialist knowledge from industries they regulate.

AI makes the problem particularly difficult because relatively few people understand frontier AI development deeply, and many of those people work for, invest in or have close connections with technology companies.

Getting sufficient expertise without creating conflicts of interest may become one of the harder institutional problems in AI governance.

What happens if AI capabilities accelerate faster than expected?

This may ultimately be the most important question.

AI policy is being created while the technology itself is changing.

If progress remains relatively gradual, governments have time to develop institutions, standards and laws. If capability development accelerates significantly, policy may need to adapt much faster.

Consider systems capable of conducting substantial scientific research, performing sophisticated cyber operations, controlling large numbers of other agents, developing significant portions of their successors, operating physical laboratories, or completing complex projects with little human intervention.

Those systems create very different governance questions from today's consumer chatbots.

The eventual AI Force may therefore need to govern technology considerably more capable than the systems available when it was created.

And then there is AGI

The announcement inevitably connects with the growing discussion around artificial general intelligence.

There is no universally agreed definition of AGI. Some AI leaders believe we are approaching it. Others reject the usefulness of the term, or believe current systems remain far from genuine general intelligence.

But governments do not have the luxury of waiting for everyone to agree on terminology.

If AI systems become dramatically more capable, policymakers will have to respond to the capabilities themselves regardless of whether everyone agrees to call them AGI.

That is why institutions such as the proposed AI Force could eventually matter.

The question is not simply who governs AGI. It is who governs increasingly powerful AI on the way there.

The international dimension may matter more than domestic policy

There is a deeper problem with national AI regulation.

Advanced AI does not respect borders particularly well.

A model developed in one country can be used in another. Research spreads. People move. Open models can be downloaded. Compute can be built in different jurisdictions. Companies operate internationally.

If advanced AI eventually creates risks that cross national borders, purely domestic rules may not be enough.

That raises the possibility that AI governance eventually requires something closer to international coordination.

Cybersecurity already works this way. Nuclear technology does too. Aviation has international standards. Financial systems require cross-border cooperation. Advanced AI may ultimately join that list.

The relationship between the US and China could therefore become one of the most consequential parts of the AI safety story.

What should we watch next?

The announcement itself leaves more questions than answers. The important developments now will be concrete ones.

Who becomes AI czar? Where does the AI Force sit within government? What legal authority does it have? Who joins it? Does Congress need to approve or fund parts of it? Does it have access to frontier AI models? Does it include independent safety expertise? How does it interact with existing cybersecurity and national-security agencies? What relationship does it have with OpenAI, Anthropic, Google, Meta and other frontier developers? Does it create new rules, or does it primarily promote development and enforce existing law?

And perhaps most importantly: what happens if the AI companies themselves ask the government to slow something down?

That question is no longer entirely hypothetical.

The AI safety debate is becoming a governance debate

For years, much of the discussion around advanced AI safety happened inside research laboratories, universities and specialist communities.

That is changing.

As capabilities increase, questions about AI safety are becoming questions about law, economics, national security, international competition, corporate responsibility and government power.

Trump's AI Force announcement is part of that transition.

At the moment, the AI Force is more an announced intention than a defined institution. It would be premature to judge what it will accomplish before we know what it actually is.

But the philosophy behind the announcement is already clearer.

The Trump administration wants the United States to remain at the front of AI development. It does not currently appear persuaded that broad restrictions on frontier development are the right response to AI safety concerns. And it believes government should nevertheless play some role in watching the industry and responding to harmful activity.

Whether those objectives can all coexist becomes a much harder question as AI gets more capable.

Because ultimately the debate is not simply about whether America should lead in AI. Most people involved in US AI policy already want that.

The much more difficult question is this. How fast should the race be run, what safeguards should exist along the way, and who gets to apply the brakes if something goes wrong?

What this article does not establish

It does not establish what the AI Force will do, because nothing about its structure, powers or funding has been published.

It does not establish that it will be a military body. That is Trump's analogy, not a stated legal form.

It does not establish who will be appointed as AI czar. Names in circulation are reporting of internal discussion, not appointments.

It does not establish that this announcement weakens AI oversight, or that it strengthens it. Both readings are currently available and neither is supported by published detail.

It does not establish that the administration and OpenAI are in conflict. They published two days apart, neither responded to the other, and both begin from the position that the United States should lead.

It does not establish that any system is superintelligent. Renaming the technology in government documents is a decision about words, not a measurement of capability, and no evaluation accompanied it.

It does not establish that the United States has rejected international coordination. It has declined a multilateral framework while separately proposing a bilateral incident-notification mechanism with China.

What it does establish is the direction being signalled: that the United States should keep developing quickly, and that harm should be dealt with through existing law rather than through new restrictions on the models themselves.

Sources

• Reuters: Trump says he will create AI Force, name AI czar (19/09/26)
• Axios: Trump wants an AI Force to fight doomers, and a scramble for AI czar is underway (21/09/26)
• CNN: Trump vows to create AI Force and appoint czar (19/09/26)
• The White House: Fact sheet: promoting advanced artificial intelligence innovation and security (June 2026)
• OpenAI: Building standards for the next phase of AI (21/09/26)
• Reuters: OpenAI calls for US to take lead in global efforts to develop technical standards (21/09/26)
• Reuters: Trump says US Justice Department could rein in AI companies if needed (22/09/26)
• Reuters: UN chief calls for AI curbs and an end to wars in his last assembly address (22/09/26)