Two speeches at the United Nations General Assembly, delivered hours apart on September 22, set out positions that cannot both survive contact with the same industry. Britain's prime minister, Andy Burnham, used his first address to the assembly to say the United Kingdom would lead an international effort to write standards for artificial intelligence and would put the subject at the center of its G20 presidency next year. The president of the United States, speaking to the same hall earlier the same day, rejected any international scheme to govern the technology and announced that his government would stop calling it artificial intelligence at all.
The gap between the two is not a disagreement about how much regulation is appropriate. It is a disagreement about whether the question belongs to governments.
Two speeches, hours apart, and one question neither answered
Burnham's framing was deliberately institutional. He told the assembly the UK would act as an honest broker and work toward a single set of global principles and standards for the development of frontier models, and he positioned the G20 presidency as the vehicle, with leaders due to gather in Manchester in 2027. "We've all heard the warnings, which we must heed," he said, a line that carried the structure of his argument: the risk case is established, the response should be coordinated, and the coordination should happen among elected governments rather than in the laboratories that build the systems.
Trump's answer, delivered first, was a categorical rejection rather than a competing design. He said the United States totally rejected any attempt to construct a globalist scheme to control AI and that his administration would "encourage it, not rein it in," according to the i Paper's account of the speech. He dismissed safety concerns as a hoax. He also said he wanted the technology renamed super intelligence, because, as The Journal.ie recorded it, "the use of the word 'artificial' makes intelligence fake," with American government documents to adopt the abbreviation SI from that point forward.
The renaming is easy to read as a rhetorical flourish. It is better read as a statement of jurisdiction. A technology called artificial intelligence invites questions about whether its outputs are real, who is accountable for them and which agency supervises them. A technology called super intelligence is presented as an arrival rather than a product, and arrivals are not regulated by committee.
What the UK is proposing to broker
Stripped of the stagecraft, the British position has three parts. The first is a set of shared principles for frontier model development, covering transparency and the access regulators need to evaluate what a model can do. British officials have told the industry that the government expects full visibility of new frontier models, and Burnham cited the country's AI Security Institute, which he said works closely with the United States and the leading laboratories, as the institutional base. The second is a claim about who decides: he said fundamental decisions on AI belonged with elected governments rather than with technology companies. The third is a venue. The G20 is where Britain holds the chair next year, and a communiqué is easier to obtain there than a treaty.
He also drew a distinction between coordinating rules and writing them. Burnham described the benefits of the technology in healthcare, education, economic growth and defense, and said he did not intend to be a doom-monger about it. What he asked for was a common frame within which national rules could be written, which is a smaller ask than a global regulator and a larger one than a voluntary pledge.
The comparison that hangs over the proposal is the European Union's AI Act, which took the opposite route: a single binding statute with obligations tiered by risk, phased in over several years, applying to any provider that wants access to the European market. That approach gives the EU real leverage over American laboratories, because access to the single market is a price the companies will pay. Britain, having left that market's rule-making, is proposing to coordinate standards without a market of comparable weight behind them, which is why the argument has to be made in the currency of shared principles rather than statutory text.
The 2008 comparison is the argument and also the weak point
The most revealing passage in the speech was the historical analogy. Burnham invoked the response to the financial crisis of 2008, when a Labour government in London helped lead the international effort, and described the mechanism as a set of principles that every country then implemented at home in the way that suited it. That model, he said, worked.
It did work, for banks, and the reason is worth stating precisely because it defines the difficulty here. The post-2008 framework worked because the crisis made the largest jurisdictions want the same things at the same time, because the standards were written with national supervisors who already had statutory authority over the firms in question, and because the firms were distributed across those jurisdictions rather than concentrated in one. None of the three conditions holds for frontier AI. The United States hosts the largest laboratories, so an international standard is mostly a constraint on American companies. The supervisor in the American system is not obviously any single agency, and the current administration has said it does not want the role. And the urgency that produced Basel III in two years has no equivalent here, because the risks are contested rather than visible on a balance sheet.
That asymmetry is the reason the American objection is not simply ideological. The United States Embassy in Canberra made the same argument in a different context, in a formal submission opposing an Australian law, with the observation that the platforms affected by such rules are overwhelmingly American companies operating on American soil, and that rules written elsewhere would reach speech at home. The same logic applies to standards drafted with British leadership and applied to models trained in California.
The warnings he cited have names and dates
Burnham's case for urgency rests on a specific run of industry events rather than on projections. Days before the assembly, an Anthropic safety researcher, Evan Hubinger, said he put the chance that AI could cause human extinction within a decade above ten percent. A colleague, Jacob Coxon, resigned from the company and said it was not acting responsibly about that risk. Sam Altman of OpenAI was scheduled to address the UN Security Council during the same week. Whatever one makes of any individual claim, the industry's own employees have become the most quoted source of alarm about the industry, which is an unusual posture for a sector asking to be trusted with self-governance.
The diplomatic backdrop is less ambiguous. More than twenty countries, mostly European, signed a push for binding safety measures ahead of the assembly, warning that development could outpace the ability to control it. China and the United States declined to endorse it. Burnham was criticized in London for not joining that initiative, and his answer was that his G20 process would absorb it.
That is a defensible sequencing choice and a risky one. A G20 track that takes a year to produce principles gives the countries already demanding binding measures a reason to wait, and gives the governments that oppose them a year in which nothing binds. The countries that signed the letter wanted obligations, and the British offer is coordination.
The half of the speech that does not need a treaty
One element of the address will be implemented regardless of what the G20 does. Burnham announced a National Centre for Information Defence, tasked with detecting, attributing and disrupting hostile state information operations and with identifying deepfakes that damage British interests. He said Russian agencies had used bots, fake websites and falsified newspaper articles, forged the branding of twenty-eight British organizations including universities and the BBC, amplified far-right narratives, and that there is evidence of attempts to interfere in the 2019 general election.
This is the part of the AI agenda that governments can pursue unilaterally, because it targets the use of the technology rather than its development. It also happens to be the part most likely to draw objections from the same quarter that objects to platform rules, since detection and attribution of foreign influence operations slides quickly toward questions about what domestic platforms must remove. Britain and the United States are also preparing new defense-related AI cooperation covering critical infrastructure, which is a form of alignment that survives the disagreement over standards.
What the G20 presidency can deliver by Manchester
A G20 chair sets the agenda, hosts the working groups and drafts the language, and that is roughly the extent of the power. If the British plan succeeds, the likely output is a set of principles that member governments commit to reflect in national law, a reporting mechanism, and a shared vocabulary for evaluation and incident disclosure. If it fails, it will fail in the way the 2008 analogy suggests: the largest jurisdiction declines to be bound, and the principles become a description of what Europe and its allies do separately.
The competition for the agenda is already visible. Norway and Finland have been assembling a coalition around binding rules, with more than twenty governments attached, and Burnham's answer was that his process would fold those efforts in rather than compete with them. Whether that happens is a question about who chairs which working group, and the countries that signed the letter have their own reasons to keep a separate track alive.
The announcement that matters most in the near term is not the standards. It is visibility. The claim that Britain expects full access to new frontier models is a testable proposition, and it can be tested against companies that have so far published evaluation results at their own discretion. Whether the British government can compel that access, and whether American laboratories treat a British request differently from an American one, will be known long before Manchester.
For now the two speeches stand as an unusually clear statement of where the line runs. One government has said that the rules for the most consequential technology of the decade should be agreed between states before the technology outruns them. Another has said that the technology is an advantage to be extended rather than a risk to be managed, and has changed what it calls the technology to make the point. Both cannot be right about the same industry, and the industry is currently headquartered in the country that made the second argument.
Primary sources
- Reuters, UK ready to lead global push on AI Standards, Burnham tells UN (September 22, 2026), for the speech text, the AI Security Institute references and the frontier-model visibility claim.
- The Journal.ie, Burnham wants UK to lead on developing AI safety, after Trump dismisses concerns as a "hoax" (September 22, 2026), for Trump's remarks on renaming the technology and the twenty-country binding-measures initiative.
- The i Paper, Burnham warns world over AI and sets UK on collision course with America (September 22, 2026), for Trump's rejection of a globalist scheme, the Hubinger and Coxon references, and the two leaders' meeting.
- Morning Star, Burnham and Trump clash on AI rules (September 22, 2026), for the 2008 financial crisis analogy and for how the post-crisis principles were implemented nationally.