Two Capitals, One Chamber: AI's Chiefs Ask to Be Bound

Diplomats built this council to stop wars. Wednesday it heard from a different kind of combatant. Four companies build the world's most capable AI systems, two headquartered in San Francisco and two in Beijing. All four sat before the UN Security Council on September 23 and asked outside authorities to write the rules that would bind them¹². Dario Amodei put the largest number of the day into a single sentence. His own industry, he warned, could pose a risk to "humanity as a whole"¹. Hours later, the country that funds most of that industry stood up in the same chamber and rejected the idea¹. Between those two moments sits the real story. An industry asked to be bound. A government kept its hand off the rope.

A Chamber Hears Something New

France held the Council presidency for September, and its foreign minister, Jean-Noel Barrot, chaired the session¹⁰. The format itself was narrow. Council rules classed it a high-level briefing, a lighter format than a formal meeting, held under the standing agenda item "Maintenance of international peace and security"¹⁰. A resolution stayed off the table from the start. The point was conversation, guided by four questions the French mission wrote in advance¹⁰. Four official briefers took the floor: Bengio, Altman, Amodei, and Delangue¹⁰. The guest list still made history. Sam Altman and Dario Amodei sat across from representatives of DeepSeek and Moonshot AI, the two Chinese labs most often named as OpenAI's and Anthropic's sharpest rivals². Clément Delangue, chief executive of Hugging Face, joined by a remote line that glitched through parts of his testimony⁵. DeepSeek founder Liang Wenfeng skipped the trip and sent representatives in his place². Four frontier labs, two from each capital, shared a Security Council agenda item for the first time on September 23¹².

What the Chief Executives Actually Asked For

Amodei went first. He made the size of the risk explicit. "If managed poorly, I even believe that AI could be a risk to humanity as a whole," he told the Council¹. Altman followed with a narrower complaint. His subject: who holds the pen on any future rulebook. "If AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone," he said¹. He pushed the point further, naming the shape any standard should take. "Obviously, these standards should not lock in incumbents or favor one business model over another. They must support open and closed model developers, new entrants and established labs, national authority and international cooperation," Altman said⁴. Read closely, that sentence asks for a rulebook wide enough to bind OpenAI's own rivals. It would be written by a body that answers to every government in the room, OpenAI included.

Delangue's Defense and Bengio's Borders

Clément Delangue framed the stakes in operational terms rather than philosophical ones. "We were attacked by AI, but more importantly, we defended ourselves with AI," he told the Council, describing an automated intrusion Hugging Face had already lived through¹. Yoshua Bengio spoke next. He co-chairs the UN's Independent International Scientific Panel on AI, and he placed the danger above any single government's reach, describing a threat "that none can contain alone and that does not respect the borders we defend," language that put the risk outside any one nation's jurisdiction⁵. Altman closed his own remarks with a warning about pace rather than capability. "Beating companies in a competitive race is not a reason to make rash decisions," he said⁵. Three speakers, three separate frames: humanity, geography, and speed. All three converged on the same request for an outside referee.

Washington Objects, Redmond Leans In

Michael Kratsios represented the United States at the same session. He directs the White House Office of Science and Technology Policy, and he rejected "centralised control and global governance of AI"¹⁶. His remarks tracked a line the administration has held for a year: technology policy stays a matter for individual governments to decide⁶. Brad Smith answered a different way, days into the same UN General Assembly week that framed the Council's session. Microsoft's president has pitched some version of independent AI evaluation for 7 years. This week he tied his company's name to the plan Anthropic has built its own safety pitch around⁷. Two American institutions, one government and one company, answered the same week's question in opposite directions.

The Same Day, a Different Bet

Mark Zuckerberg opened his own keynote 6,000 miles away, the same day the Council met. "It's no secret that people in the industry have a wide range of views about how the future is going to go," he told the Meta Connect crowd. Then he distanced his company from the slowdown Altman and Amodei were requesting a continent away³. Anthropic had priced its flagship Opus 5.5 model 20% cheaper than its predecessor the day before the session. It kept racing rivals on cost and speed even as its chief executive testified⁹. Two Anthropic alumni were separately pricing a startup built on recursive self-improvement, the exact capability Amodei named as a risk to humanity. Its price: $5 billion, the same week⁸. Every company asking for restraint kept shipping at full speed. Read as strategy, the testimony functions as insurance bought in public while the business runs at its usual pace in private.

Testimony Stands Alone

A resolution needs the consent of five veto-holding members, the United States and China among them. Wednesday's format skipped that step entirely, by design¹⁰. China's position still surfaced. Beijing's delegation told the Council it supports international standards and a central role for the UN in global AI governance. That stance sits directly opposite the American one, delivered in the same chamber on the same day¹⁰. Russia added a third view. Its delegation questioned whether AI, as a broad subject, falls inside the Council's mandate at all. The topic, Moscow argued, belongs in more specialized forums instead¹⁰. Three permanent members, three separate positions, on a body built to need agreement among all five before it can act. That split keeps a binding rule a distant prospect, whatever the chief executives asked for⁶. Read plainly, the executives requesting rules understand the difference between a hearing and a vote better than anyone else in the room.

By the numbers

• 15: member states seated on the Security Council that heard Wednesday's testimony¹.
• 4: frontier AI labs represented in the chamber, two American and two Chinese, the first time rival powers' labs shared the agenda¹².
• $5 billion: the in-talks valuation for Mirendil, a recursive-self-improvement startup founded by two Anthropic alumni, priced the same week Amodei named that exact risk⁸.
• $200 million: Mirendil's seed round three months earlier, the marker its new valuation runs five times past⁸.
• 7: years Microsoft President Brad Smith says he has pitched independent AI evaluators, the plan Anthropic now centers its own safety pitch on⁷.
• 20%: the price cut Anthropic gave its own flagship model the day before its chief executive asked diplomats for restraint⁹.
• 5: permanent Security Council members whose consent a binding resolution would need; Wednesday's briefing format skipped that requirement by design¹⁰.

What to watch

Watch whether the Security Council schedules a follow-up session, or whether September 23 stands alone on the calendar¹. A briefing stays a one-time event unless the Council chooses otherwise. Kleiner Perkins could close the Mirendil funding round within weeks. That timing offers the clearest signal yet of how seriously investors price the exact risk Amodei named⁸. Brad Smith's embedded-evaluator pitch still needs a lab willing to host one, before it becomes more than a floated idea⁷. The next real test of Washington's position arrives whenever the Council puts AI back on its agenda, and whether Beijing's delegation returns with the same argument.

Sources

1. Al Jazeera, "AI corporate leaders tell UN the industry needs global regulation," Al Jazeera, Sept. 24, 2026, https://www.aljazeera.com/news/2026/9/24/ai-corporate-leaders-tell-un-the-industry-needs-global-regulation
2. Seoul Economic Daily, "DeepSeek, Moonshot AI Join U.N. Security Council AI Session," Seoul Economic Daily, Sept. 23, 2026, https://en.sedaily.com/international/2026/09/23/moonshot-deepseek-to-join-openai-anthropic-at-un-talks
3. The Associated Press, "Meta bets big on wearable AI, new AI agent as Zuckerberg pushes back against AI doom," The Associated Press, Sept. 23, 2026, https://www.click2houston.com/business/2026/09/24/meta-bets-big-on-wearable-ai-new-ai-agent-as-zuckerberg-pushes-back-against-ai-doom/
4. PYMNTS, "Altman and Amodei Call for Global AI Guardrails at UN," PYMNTS, Sept. 23, 2026, https://www.pymnts.com/news/artificial-intelligence/2026/altman-and-amodei-call-for-global-ai-guardrails-at-un/
5. ABC News, "OpenAI, Anthropic CEOs at UN call for global cooperation on AI: 'We are at a crossroads,'" ABC News, Sept. 23, 2026, https://abcnews.com/Politics/openai-anthropic-ceos-call-global-cooperation-ai-crossroads/story?id=136697471
6. FourWeekMBA, "OpenAI, Anthropic, and the UN Security Council: What a Record Produces and What It Does Not," FourWeekMBA, Sept. 24, 2026, https://fourweekmba.com/ai-openai-anthropic-un-security-council-ai-governance-record-vs/
7. Bloomberg, "Microsoft President Backs Deployment of AI Safety Evaluators," Bloomberg, Sept. 23, 2026, https://www.bloomberg.com/news/articles/2026-09-23/microsoft-president-backs-deployment-of-ai-safety-evaluators
8. Bloomberg, "Ex-Anthropic Staffers' Self-Improving AI Startup in Talks to Raise at $5 Billion Value," Bloomberg, Sept. 22, 2026, https://www.bloomberg.com/news/articles/2026-09-22/ex-anthropic-staffers-ai-startup-in-talks-to-raise-at-5-billion-value
9. VentureBeat, "Anthropic releases Claude Opus 5.5, beating Fable 5.1 on key agentic benchmarks at 60% cheaper API price," VentureBeat, Sept. 22, 2026, https://venturebeat.com/technology/anthropic-releases-claude-opus-5-5-beating-fable-5-1-on-key-agentic-benchmarks-at-60-cheaper-api-price