NextFin News - Sam Altman has thrown OpenAI's weight behind a US-led international forum to set global artificial-intelligence safety standards, arguing that the technology is now too consequential to be governed by any single country — or by the companies building it. In an essay published Wednesday, the OpenAI Group PBC chief executive proposed a body that would "establish accepted standards, provide expert and impartial analysis of capabilities and risks and make the technology available to nations and companies that participate and follow the rules." The pitch, which follows a closed-door session between AI chiefs and G7 heads of state in France, marks the clearest signal yet that frontier labs want a seat at the table before binding rules exist — and that Washington, not Brussels or Beijing, should hold the pen.
The Situation: A Forum With American Leadership
Altman's proposal calls for a forum bringing together government representatives and independent technical experts. It would, he wrote, "serve as a governance mechanism over the labs and guard against the commercial pressures that can lead to unsafe racing." As models, he pointed to global aviation safety rules, international financial standards and the International Atomic Energy Agency, created in 1957 to police civilian nuclear energy even while the United States and Soviet Union built their arsenals. His argument is that rival powers have found ways to govern dangerous technologies together.
"Everyone on Earth should benefit from this technology and determine for themselves how best to use it," Altman wrote — a line that casts the forum not just as a safety body but as a distribution mechanism for AI's gains.
The essay follows the G7 summit in Évian-les-Bains, France, where on June 17 AI executives sat down with the leaders of the world's richest democracies for a working lunch on "ensuring a safe, rapid, and effective deployment of artificial intelligence." Anthropic chief executive Dario Amodei and Google DeepMind's Demis Hassabis both argued that the United States should lead coordination on AI rules while working with allied nations. Altman attended alongside roughly a dozen other technology executives — including Mistral AI's Arthur Mensch, Meta's Alexandr Wang, Salesforce's Marc Benioff, Cohere's Aidan Gomez, Black Forest Labs' Robin Rombach and Synthesia's Victor Riparbelli.
At the gathering, Altman told leaders: "The technology's future must be shaped by people, democratic institutions and society as a whole, not just by the companies building the most capable systems." He also warned them: "Do not cede your responsibilities to AI labs like mine."
According to people familiar with the discussions, Amodei called on countries to "resist the temptation to splinter" and proposed cooperation on structured access to frontier models and trade in chips and critical components among allied nations — while excluding China. Canadian Prime Minister Mark Carney agreed the US could play a lead role, though he declined to comment publicly. Indian Prime Minister Narendra Modi, meanwhile, raised concerns about export restrictions, stressing that democratic nations need access to advanced AI tools to protect critical infrastructure.
Why the Labs Want a Standards Body — and Why Washington
The mechanics here are straightforward, and they run in both directions. For the labs, divergent national rules are a business problem: if Europe, Canada, Japan and the United States each write their own AI law, the homogeneous global market the companies need to amortise billion-dollar training runs dissolves into compliance fragments. Amodei's plea not to "splinter" is, in commercial terms, a plea for one rulebook.
But a US-led forum also solves a political problem for the companies. By inviting governments to write the standards, the labs pre-empt harsher, slower, legislature-driven regulation — and they get to help draft the rules that will govern them. OpenAI said earlier this month that it expected the summit to produce a package of voluntary commitments on youth safety and frontier risks in cyber and bio; the companies want those voluntary pledges to become the de facto global baseline before any binding treaty locks into place.
Choosing Washington as the anchor is not accidental. The United States is home to the most capable frontier models and the deepest capital markets for AI infrastructure. It also holds the leverage: days before the summit, the Trump administration imposed export controls on Anthropic's newest models, Mythos 5 and Fable 5, citing national security, briefly forcing the company to pull them from foreign users. When the state can switch off cross-border access to your product, aligning with that state's governance architecture is rational self-preservation. Anthropic restored access to Fable 5 this week, but the episode made the hierarchy unmistakable.
The geopolitical framing is explicit in the design. A coalition that coordinates chip trade and frontier-model access among allied nations while shutting China out functions less like a technical standards body and more like a security alliance — a kind of "NATO for AI," in the description of policy commentators, with Beijing outside the tent. That may be defensible on national-security grounds; it is not a neutral safety outcome, and it will be read as containment by the excluded party.
Cyclical or Structural: This Is a Regime Shift, Not a Moment
The right read on this episode is structural, not cyclical. A cyclical interpretation would treat the G7 forum push as a one-off reaction to the Anthropic export-control shock — a temporary alignment that reverts once the immediate regulatory pressure eases. That underestimates the shift underneath it.
Three things have changed for good. First, the locus of AI power has moved: a handful of private companies now control cognitive capabilities that governments, businesses and citizens depend on, and they arrived at the G7 table not as petitioners but as indispensable partners. Second, national security has supplanted safety as the organising principle of AI governance; the proposed coalition would function as a security alliance rather than a purely technical body. Third, the United States has demonstrated it will weaponise access to frontier models as an instrument of statecraft, which means governance and geopolitics are now the same conversation.
None of these is mean-reverting. Export controls do not get un-invented; a security framing does not unwind back into a purely technical discussion; and once chief executives have sat alongside heads of state as co-governors, they do not return to being mere vendors. The forum proposal is the institutional expression of that new order. Europe and Canada, for their part, have embarked on their own "AI sovereignty" efforts to reduce dependence on US technology firms — a reaction that confirms the shift rather than reversing it.
The Second-Order Problem: Standards as a Moat and a Wall
The first-order story is safety coordination. The second-order story is market structure. A US-led standards regime that makes technology available only to "nations and companies that participate and follow the rules" creates a club good: membership confers access, and the incumbents help decide the membership criteria. That is a moat dressed as a safeguard.
The valuations make the stakes concrete. OpenAI closed a $122 billion funding round in March 2026 at an $852 billion post-money valuation, by far the largest private raise in the sector's history. Analysts estimate that to justify that price, the company would need to generate roughly $95 billion to $105 billion in free cash flow by 2030. Its first-quarter revenue came in at $5.7 billion, but at a negative 122% adjusted operating margin the company spent about $2.22 for every dollar earned — meaning the economics depend on scaling far beyond today's revenue base.
Anthropic, by contrast, is pursuing a parallel listing with higher estimated run-rate revenue, a faster path to profitability and what analysts put at a 40% enterprise market share versus OpenAI's 27%. In a world where the standard-setting club is chaired by the same governments that award compute-export licences and approve model rollouts, the firms closest to Washington gain a structural advantage that no amount of open-source competition can easily erode. The rulebook becomes part of the competitive landscape — and the people writing it sit in the same room as the companies that benefit most from it.
The exclusion of China is the wall half of the same design. Formalising the bifurcation of the AI world into two technology spheres is a geopolitical choice with commercial consequences: it protects allied supply chains while raising the cost of serving any market that touches the excluded bloc. For chipmakers, cloud providers and model developers with global ambitions, that is a new and durable constraint on where they can sell and to whom.
The Counter-Thesis: Enforcement Is the Open Question
The strongest objection is also the simplest: unlike aircraft and nuclear enrichment plants, frontier models are trained inside data centres with little outside visibility. Inspectors can walk a reactor floor; they cannot easily tell whether a lab is racing ahead in secret or quietly exceeding agreed capability thresholds. A voluntary forum without verification is a promise without a police force.
Tom Wheeler, a senior fellow at the Brookings Institution, argued this week that the G7 should accept the industry's offer to help set standards — but only "on the condition that any resulting agreement is made enforceable rather than voluntary." That is the hinge the whole proposal turns on. Altman's own citation of the IAEA cuts against him: the nuclear agency works because states consented to inspections with teeth. A standards body that can neither inspect training runs nor sanction violators would be, at best, a talking shop — and at worst, a fig leaf that lets the labs claim oversight while racing anyway.
There is also a divergence within the camp. Amodei has argued for rules closer to the Federal Aviation Administration — a prescriptive, permit-based model — which is a harder regime than the standards-setting body Altman describes. If the two labs cannot agree on how strict the rulebook should be, a global consensus among seven governments and a dozen companies looks harder still.
The falsifying signal is specific: if, within 12 months, the G7 produces a signed framework with independent inspection authority over training runs and a real sanction mechanism — and at least one major lab submits to unannounced audits — then the forum is becoming a genuine governance regime. If instead the outcome is a voluntary charter with no verification, the structural reading above is wrong and this was mostly positioning.
Who Benefits, Who Is Exposed
The winners from a US-led club are legible. American frontier labs gain a rulebook they helped write and a home regulator whose export-control power they can anticipate rather than fear. US chip and cloud incumbents benefit from a coalition that keeps advanced compute inside allied borders. Governments inside the tent gain a channel to the companies that control the technology without having to build state capacity from scratch.
The exposed are equally clear. European and Canadian sovereignty efforts face a harder path if the de facto baseline is set in Washington; firms caught between US rules and local autonomy demands will carry the compliance cost. Open-source and non-aligned developers risk being locked out of the certified distribution channels the forum would create. And China, excluded by design, accelerates its parallel build-out — which is precisely the dynamic the forum is meant to contain but may instead entrench.
What Comes Next
In the short term, expect a voluntary commitments package — youth safety, cyber and bio frontier-risk pledges — to be announced as the summit's deliverable, with the forum itself remaining an idea rather than an institution. Medium term, the battle shifts to whether enforcement gets written into the framework; that is where the demand for enforceability and the labs' preference for flexibility collide. Long term, the question is whether the US-led club becomes the de facto global baseline or fractures under pressure from Europe's sovereignty push, India's access demands and China's parallel build-out.
Base case: a voluntary forum launches with US backing, China stays outside, and enforcement remains contested. Upside case: the G7 agrees to an enforceable inspection mechanism, giving the model real teeth. Downside case: the proposal stalls on verification, Europe accelerates its own sovereignty stack, and the AI world splits into competing regulatory blocs regardless.
Altman's forum is being sold as a safety mechanism; it will function as a map of power. The labs get a rulebook they helped write, Washington keeps the pen, and the countries outside the club learn what it costs to be excluded. That is not a neutral outcome — it is the point.
Explore more exclusive insights at nextfin.ai.
