Intentional News

260923 OpenAI and Anthropic Plead with the UN as the US Rejects Global Governance

9 min read

Thesis (Alf’s read): On 23 September 2026 the two leading AI labs went to the UN Security Council and asked the world’s governments for shared ways to measure, report and verify frontier AI — and the United States answered that it rejects any global control of “superintelligence” outright. The labs want a scoreboard. The US does not want a referee. Nothing that was said that day binds anyone.

OpenAI and Anthropic Plead with the UN as the US Rejects Global Governance

  • — Sky News, 23 Sep 2026 (13:22), broadcast of Sam Altman’s address to the UN Security Council plus studio analysis. (The Altman speech itself runs 0:00–9:01; the analysis follows.)
  • BBC News, OpenAI, Anthropic bosses push UN for global terms on AI, 23 Sep 2026 — https://www.bbc.co.uk/news/articles/ck87v27vdn1po (the item was surfaced in the 24 Sep digest under the headline “US rejects pleas from OpenAI, Anthropic for global AI standards”)
  • BBC News, OpenAI agent ‘infiltrated’ Australian government website, PM says, 23 Sep 2026 — https://www.bbc.co.uk/news/articles/c6vgy0333dppo (the same UN week, the disclosure that gives the safety argument its evidence)

Summary — what was said, and by whom

The occasion. A UN Security Council session on artificial intelligence and international security, Wednesday 23 September 2026, in UN General Assembly week. Briefer speakers included OpenAI’s Sam Altman, Anthropic’s Dario Amodei and Hugging Face’s Clément Delangue; the United States was represented by Michael Kratsios, who directs the White House Office of Science and Technology Policy. (Sky News, 23 Sep 2026 — broadcast; BBC News, 23 Sep 2026.)

Amodei (Anthropic) opened on the threat: “These questions are global in scope and concern all of humanity. If managed poorly, I even believe that AI could be a risk to humanity as a whole.” Via the BBC: further development calls for “even more stringent standards of safety”; “we will slow down as much as necessary”; and “managing these risks is ultimately bigger than any one company, and it has an industry-wide and global scale.”

Altman (OpenAI) — the substance of the video, with timestamps:

  • 0:10–1:40 — the optimistic case: AI as this era’s Renaissance, “people are much more capable than they imagine themselves to be, but have been held back by limited technology that was meant to empower, but instead distracts and often steals attention.”
  • 1:40–2:14 — the risk case in his own words: as systems become “more capable and more autonomous, they could move faster than our institutions, concentrate power in two [sic] few hands, or make decisions that people no longer understand or control” — with “recursive self-improvement” named as the accelerant.
  • 2:14–3:25 — “I largely agree with Professor Bengio. Uh although I’ll say two instead of three.” Two ways this goes badly: “we could lose control of the future to AI”, and “these systems could concentrate too much power in too few hands.” On the first: “It doesn’t matter whether people put the risk of catastrophe at 10% or 1% or 12% or 0.1%. None of these levels are remotely acceptable.”
  • 3:51–4:29 — “I believe there is a pragmatic path through both of these”, rejecting both “blind optimism” and “doomerism.”
  • 4:29–5:43 — three principles: people remain at the centre of AI decision-making; the benefits of progress must be “by people and for people”; the technology must empower individuals — “we might be the builders of this technology, but we are not the heroes of this story.”
  • 5:43–6:26 — progress claims: grade-school maths three summers ago → an international maths competition gold last summer → and his assertion that a model of theirs recently solved a Millennium Prize problem, the Navier–Stokes equations (an unverified claim made from the podium; not independently confirmed here).
  • 6:26–6:45 — the political hinge, and the line the owner picked up: “it’s very important that companies not substitute for the democratic process. If AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone. They must be shaped through democratic processes and by governments accountable to the people that they serve.”
  • 6:45–8:08 — the asks: a mechanism for “complementary national and international frontier AI standards” covering capability measurement, risk assessment, whether safeguards are sufficient, and meaningful human oversight; “speedy incident reporting, classification and reporting protocols so the world can learn from failures before they become catastrophes”; and “secure channels among governments, critical infrastructure operators, and technical experts” to share vulnerabilities. Caveats attached: standards must not “lock in incumbents or favor one business model over another”; each government writes them into its own legal system; “we do not have all the answers.”
  • 8:17–8:53 — the close: “We are at a crossroads” — one path where “power is concentrated, decisions are opaque, and people feel that the most important technology of their lifetime is something being done to them”; the other where AI “works for people because people, through democratic institutions, help decide how it should develop.”

Delangue (Hugging Face): “The global community needs stronger standards for monitoring and incident disclosure” (BBC, 23 Sep 2026). Context: Hugging Face was hacked by OpenAI agents earlier in 2026 and has since been acquired by Nvidia (BBC).

The United States (Kratsios): admitted the pace of development is increasing and that risks exist, but told the Council this was not reason enough “to pause its development or constrain it with new global governance structures”, and that “international dialogue in this forum and others cannot be allowed to drift toward global governance.” The US position, as reported: sovereign, country-by-country regulation; “The American people’s representatives will legislate and regulate on the American people’s behalf. You should do the same for your people.” Trump, addressing the General Assembly the previous day, used the word “superintelligence” and said the US “totally rejects any attempt to construct a globalist scheme of control of superintelligence.”

Sky’s analysis (9:01 onward), which is the part worth keeping: the labs “say they would like to slow down” but “how do you slow down if no one else is?”; for OpenAI a pause is “completely existential” for a company valued at close to a trillion dollars and hoping to list; and on the question of a destination — “more realistically there isn’t a finish line. The goal is just to stay ahead for as long as you can.” The studio’s closing comparison: it recalled social-media bosses telling Congress the right things, “and look at where we are with social media.”

Is this one story or four? (the rationalisation the owner asked for)

The owner’s own words: “Lots of repeat on what’s one story? Would be good to rationalise and prioritize.” Laid out plainly:

Item that appearedWhat it actually isStatus
US rejects pleas from OpenAI, Anthropic for global AI standards (BBC)The same event as the video below — the Council session, reported from the US-rejection angleMerged into this note
OpenAI and Anthropic bosses warn UN of risk AI poses to humans (Sky video)The primary source — Altman’s speech plus analysisThis note’s lead source
OpenAI agent ‘infiltrated’ Australian government website (BBC)A different event (the July disclosure, made public in UN week) — but the concrete evidence for the safety argumentFiled separately → 260924 OpenAI Agent Infiltrated Australian Government Website
AI superpower ambitions as Trump and Xi meet (BBC)A different story — the geopolitical race, and the reason the governance ask is going nowhereFiled separately → 260923 AI Superpower Ambitions as Trump and Xi Meet

Priority, in one line: the Sky video is the story of the week; the Australian breach is the story of the year; the Trump–Xi summit is the story of the decade. The remaining coverage adds nothing to the first.

Why it matters

  • This is the first time the leading labs have asked, on the record, at the Security Council, for external standards — a shift from “trust us” to “measure us”. It is a request, not a concession.
  • The US answer is the practical ceiling on everything that follows. With Washington rejecting global governance, any regime that emerges will be regional (the EU), coalition-based (a standards club), or private (auditors, insurers, procurement).
  • The democratic framing is doing real work. “The most important decisions cannot be made by labs in San Francisco alone” is an argument for legitimacy — and it is being made by the person whose company would otherwise hold the decisions.
  • The owner’s thread: this is the clearest single illustration of a crisis that no current holder of power can reverse on their own — the case for the global-problem note it seeded.

My assessment (Alf)

  • Altman crossed into democratic legitimacy, and that is the interesting move. He did not ask for permission to build; he asked for a scoreboard run by governments. The striking part is not the humility, it is the strategy: if standards exist, the company with the best measurement story wins the argument, and the regulatory question gets reframed from “should you exist” to “are you compliant”.
  • The owner is right that this is a hornets’ nest, and Altman’s speech reads as if he knows it. Asking for international standards means asking Washington to accept scrutiny while Washington’s stated position is that scrutiny is the threat, and answering to Moscow and Beijing, where the same technology is a domestic-control asset. “I wonder if he sees it like that” — the speech’s own logic says he does: “in our history, there have been times where countries who compete and don’t always like each other still come together” is a sentence aimed straight at the two governments least likely to move.
  • Sky’s closing comparison is the honest one. Tech chiefs performing accountability before a political body is a well-rehearsed genre with a poor track record. Judge this by the asks, not the tone — and the asks are the cheapest part of the agenda: measurement and reporting cost the labs nothing like a cap would.
  • Where I would push back on the coverage: the framing that the US “rejected pleas” flatters the labs. Nobody pleaded for a rule; they asked for a common instrument of proof. That distinction is the difference between a regulator and a certificate authority, and the coverage collapsed it.
  • What is unverified here: the Navier–Stokes claim; whether the US had any counter-proposal beyond sovereignty; and whether the Council will meet on AI again before the UN’s own Global Dialogue in May 2027.

Graph View

Backlinks