Three weeks before he stood in front of the world's most powerful diplomatic body warning that humanity could "lose control of the future of AI," Sam Altman was on a stage in North Carolina backing the exact opposite policy. That whiplash is the real story behind the UN Security Council AI regulation push that OpenAI's Altman and Anthropic's Dario Amodei made on September 23-24, 2026 — a plea for global standards that arrived at a venue built for war and peace, not technology policy, and that landed just weeks after Altman himself had championed a US-backed framework built to keep regulators out of AI's way.
What Altman and Amodei Actually Asked For
The session, held during UN General Assembly week in New York, was reportedly the Security Council's first-ever high-level briefing dedicated specifically to AI safety risks. Altman and Amodei didn't share a stage — Altman spoke in person, Amodei by video — but their asks rhymed more than they diverged.
Altman's "Middle Path"
Altman framed his pitch as a middle path between doing nothing and shutting AI down. It rested on three principles: AI systems must stay under human control, must benefit people broadly, and must "empower people individually" rather than concentrate power. Concretely, he called for international standards for measuring AI capabilities and assessing risk, for determining when safeguards are sufficient, for preserving human oversight, and for building secure government-to-government channels to share intelligence on emerging AI threats. He also pushed for fast, accurate incident reporting, arguing the world needs to "learn from failures before they become catastrophes" — and argued that decisions about AI's future should sit with democratic institutions "accountable to the people they serve," not with "labs in San Francisco alone."
Amodei's Three Mechanisms
Amodei's video remarks were more mechanical and, notably, more alarming in tone: he said that "if managed poorly, I even believe AI could be a risk to humanity" as a whole. His ask broke into three pieces — narrow global agreements on specific dangers, such as a ban on using AI to help design biological weapons; evaluation and verification systems so countries can check whether rivals are honoring their commitments; and common global testing standards paired with a notification system for AI security incidents, modeled loosely on how nations already report other kinds of hazards. Turing Award winner Yoshua Bengio, also briefing the Council, called the dangers "real and imminent," while Hugging Face CEO Clement Delangue described the more mundane reality already playing out: companies defending against AI-driven attacks using AI tools of their own.
| Proposal | Altman (OpenAI) | Amodei (Anthropic) | US position (Carolina Principles) |
|---|---|---|---|
| Core ask | Global standards for capability testing, risk assessment, human oversight | Narrow bans (e.g. bioweapons), verification systems, incident notification | Sector-specific rules, no new regulatory agencies |
| Who decides | "Democratic institutions," not labs | National governments, mutually verified | Individual nations, sovereignty-first |
| Binding? | Not specified as binding | Framed as agreements between states | Explicitly non-binding accord |
| Delivered where | UN Security Council, Sept. 23, 2026 | UN Security Council, Sept. 23, 2026 (via video) | G20 summit, North Carolina, Sept. 3, 2026 |
The Whiplash: Altman Backed the Opposite Framework Three Weeks Earlier
Here's the part the UN briefing coverage tends to skip: on September 3, 2026, at a G20 tech and innovation summit in North Carolina, Commerce Secretary Howard Lutnick and White House Office of Science and Technology Policy Director Michael Kratsios announced that all 20 G20 nations — including China and Russia — had unanimously endorsed a US-backed framework now known as the Carolina Principles. It pushes member states toward sector-specific AI rules, away from new regulatory agencies, and toward close collaboration with industry on evaluating new technology. It's expected to go to formal adoption at the G20 leaders summit in December at Trump National Doral in Florida.
Who Else Signed On, And Why
Altman was there supporting it, framing AI adoption as inevitable and comparing it to the rollout of electricity. He wasn't alone: Nvidia's Jensen Huang argued regulators should target "practical and actual harm, and not theoretical and hypothetical harm," calling AI "the great equalizer." Elon Musk warned that over-regulation could choke off AI's potential to expand the global economy by 20 to 30 percent. Palantir's Alex Karp pushed back on what he called "doom messaging." Anthropic wasn't absent from that room either — the company's Tom Brown was there pressing for streamlined data center permitting, a very different kind of policy ask than Amodei's UN warning about existential risk.
Three weeks is not a long time to pivot from "let industry evaluate itself" to "the world needs binding international testing standards." That gap is exactly what's fueling the skepticism that this was less a safety plea and more a competitive maneuver — a way for the two best-funded labs in the industry to shape whatever global rulebook eventually gets written, while leaving the door open to keep operating under the lighter domestic framework they helped write at home.
Why the Security Council, of All Places?
The UN already has AI-specific machinery. A 2024 General Assembly resolution, nonbinding, called for monitoring AI risks and protecting data. The Assembly has also floated two new bodies aimed squarely at this problem: an Independent International Scientific Panel on AI and a Global Dialogue on AI Governance. None of that requires the Security Council, whose fifteen members exist to handle threats to international peace and security — sanctions regimes, peacekeeping mandates, ceasefire resolutions — not technology standards.
What a Council Session Actually Produces
One structural read of the briefing, floated by analysis site FourWeekMBA, draws a sharp line between what the Security Council produced here and what it's built to produce: "a record is citable; an obligation is enforceable." A high-level briefing generates a transcript and headlines, not a resolution — the Council didn't vote on anything. If Altman and Amodei were hoping to plant the seed for something Chapter VII-grade and binding down the line, September's session wasn't that. It was a hearing, not a treaty.
The Pushback on Hacker News
The venue choice didn't sit well with the thread that formed around the story on Hacker News. Commenter impossiblefork called it "a really strange decision to choose the security council as the forum, given its purpose." Others went straight at motive: yesco summarized the ask as "tech leaders demand nations of the world give them a moat or else," while JackSlateur mocked the framing outright — "please! please! stop ignoring me! look at me, look at me! I'll set the house on fire if you do not pay attention to me! also, give me all your money!" Commenter soraminazuki pointed to the underlying contradiction: companies citing catastrophic danger as justification while continuing to build and ship the very systems they're warning about, rather than slowing down. watwut took that further, arguing accountability should come from liability, not self-authored rules — "just stop shipping and building it if it is so dangerous" — and drew a comparison to food safety, where companies answer for harm after the fact rather than writing the regulatory playbook themselves.
The US Says No, Directly to the Council
The clearest answer to "why this venue" may simply be that Washington chose to answer there. The day after Altman and Amodei spoke, September 24, Kratsios told the Council: "We totally reject all efforts by international bodies to assert centralised control and global governance of AI." He argued the path to a flourishing AI-powered future runs through "the independence and sovereignty of nations," not "bureaucratic management," and told member states plainly: "The American people's representatives will legislate and regulate on the American people's behalf. You should do the same for your people." On social media the same day, Kratsios went further, dismissing "ideological fixations on social equity, climate catastrophism, and so-called existential risk" as "dangers to progress."
Two Governments, One Room, Opposite Instructions
That's the odder split hiding in plain sight: the same US administration that pushed the light-touch Carolina Principles through the G20 less than a month earlier sent its own science and technology chief to the Security Council to tell the world, on the record, that it won't accept exactly the kind of international standards its own AI companies had just asked for. Altman and Amodei weren't arguing with a rival government's position — they were arguing, functionally, with their own.
What to Watch Next
Nothing binding came out of New York, and nothing was expected to. The near-term test is the G20 leaders summit in December at Trump National Doral, where the Carolina Principles are due for formal adoption — a much more concrete marker than a Security Council briefing, since it involves an actual signature rather than a warning speech. Watch whether Altman and Amodei show up there with the same urgency they brought to the UN, or whether the global-standards language quietly stays confined to venues that can't actually bind anyone.
-EditorZ
Photo by Matthew TenBruggencate on Unsplash
Post a Comment