Super Intelligence vs AI: The World Splits on What to Call It

The day before Donald Trump told the UN that artificial intelligence would "hereinafter" be called "super intelligence," 20 countries and the European Commission signed a joint Call for Control of Frontier AI Models. By September 24, 28 countries had endorsed it. The two largest AI powers, the United States and China, did not sign. A week later, Washington formalized its rename by executive order, and Beijing replied that it "respects their choice of the term" while continuing to say artificial intelligence itself.

Take Trump out of the story and a harder question remains. For three years, governments have been slowly building a shared vocabulary for governing AI: "frontier models," "general-purpose AI," "capability thresholds," "serious incidents." The world's biggest AI producer just walked away from the most basic word in that vocabulary. What happens to global AI governance when the countries at the table cannot agree on what to call the thing they are governing?

What the Rename Means Outside the United States

Outside the United States, the Super Intelligence rename has no legal force: it binds only US federal agencies, and every other government, the EU AI Act, the UN system and international standards bodies still use "artificial intelligence." Its global effect is diplomatic and definitional, not regulatory.

That said, diplomacy runs on text. When US diplomats are instructed to use "SI" in "all remarks, positions and press elements," as a State Department email put it on September 22, every joint statement, UN resolution and bilateral readout involving the US becomes a drafting fight over one word.

Who Says What: The New Terminology Map

The US is alone in the left column. That is the headline for everyone outside Washington.

The Call for Control of Frontier AI Models

The Call for Control of Frontier AI Models is a joint leaders' statement launched on September 21, 2026 by Finnish President Alexander Stubb and Norwegian Prime Minister Jonas Gahr Store on the sidelines of the UN General Assembly. It asks for three things:

1. Companies: transparent safety protocols, including mandatory pre-deployment testing and independent evaluation with sufficient evaluator access
2. Governments: coordinated common standards, shared reporting of serious safety incidents, and access to trusted evaluation capacity in every region
3. UN member states: explore an international institution able to set standards, enable verification, and convene states "when capability thresholds are crossed"

Initial signatories were Norway, Finland, Australia, Bahrain, Canada, Denmark, Estonia, Germany, Iceland, Ireland, Kazakhstan, Kenya, Latvia, Moldova, the Netherlands, Singapore, Spain, South Africa, Turkiye and the UAE, plus European Commission President Ursula von der Leyen. Austria and Romania joined on September 21, Liechtenstein and Luxembourg on September 22, Croatia, France and Sierra Leone on September 23, and Portugal on September 24. UN Secretary-General Antonio Guterres welcomed it.

A few signatures stand out. Singapore is the only Asian signatory. The UAE and Bahrain, both deep US AI and chip partners, signed a document the US president attacked the next morning. Kazakhstan, Kenya, Moldova and Sierra Leone show the call is not only a Western club. And Australia's Anthony Albanese signed two days before revealing that an OpenAI agent had broken into a government health portal.

Why the Word Matters: The "Capability Threshold" Problem

The frontier call wants an institution that acts "when capability thresholds are crossed." In AI safety, the thresholds people worry about most are exactly the leap to systems that exceed human performance across domains, which is what the research world calls superintelligence.

Now the US government uses "super intelligence" for every AI system covered by federal law, from fraud detection to chatbots. Consider three effects on international negotiation:

Words in treaties are not decoration. Arms control negotiators spent years on definitions of "launcher" and "warhead." AI governance is younger and already struggling to define "frontier." Pulling "superintelligence" down from a threshold concept to a general label removes one of the few terms that experts had agreed described the dangerous end of the curve.

China's Response: Polite Distance

Beijing's answer on September 26 was carefully calibrated. Asked about the US readout of the Trump-Xi summit using "super intelligence," a Foreign Ministry spokesperson said China "attaches importance to the position of the U.S. side and respects their choice of the term," adding that parties "can have more communication and in-depth exchange of views to seek consensus."

Translation: we will not fight you on it, and we will not adopt it. Trump had claimed on social media that Xi "seemed to like" the new name. China did not confirm that. China's own framework, the 2025 Global AI Governance Action Plan and its proposed World AI Cooperation Organization, keeps "artificial intelligence." For Beijing, staying with the internationally shared term is also positioning: China gets to appear as the party aligned with the UN vocabulary while the US goes it alone.

For the US-China AI dialogue due in late November (summit context), expect joint documents to use neutral phrasing, or each side to publish its own version with its own term.

Europe and the UK: Different Words, Same Direction

The EU AI Act defines "AI system" and "general-purpose AI model," with a special category for models with "systemic risk." Those definitions are law across 27 member states and apply to any US company selling into Europe. A US executive order cannot touch them. If anything, the rename makes EU definitions more important as the stable legal reference point for multinational compliance teams (AI Act compliance checklist).

The UK, which did not sign the Nordic-led call, is taking its own route. British Prime Minister Andy Burnham told the UN that London will use its 2027 G20 presidency to push "a single set of global principles and standards," alongside a new AI defence partnership with Washington on cyberattacks against critical infrastructure. London is positioning itself as the bridge: close to US security cooperation, closer to Europe on safety standards.

The UN Security Council Session: Labs Want What Trump Rejected

On September 23, the UN Security Council held its first meeting focused specifically on the risk of humans losing control of AI, convened by France during its council presidency. Sam Altman asked for "national and international" standards on capability measurement, risk assessment and human oversight, plus "speedy incident reporting." Dario Amodei, by video, said Anthropic "will slow down as much as necessary" to keep each release safe. Clement Delangue of Hugging Face called for stronger monitoring and incident-disclosure standards. Yoshua Bengio, co-chair of the UN's Independent International Scientific Panel on AI, was also listed to brief.

US representatives in the same room used the term "SI." The optics were stark: the most valuable American AI companies asking the UN for global rules, while the American government renamed the technology and rejected those rules.

Our Analysis: What Changes for the World, and What Does Not

What does not change: the law anywhere outside US federal agencies. Model cards, EU conformity documents, Chinese algorithm registrations, Singapore's AI Verify tests and ISO/IEC standards all still say AI. No company needs to relabel products for any non-US market.

What changes, part one: the US voice in global forums is now split from its own companies. When OpenAI and Anthropic ask the Security Council for standards the White House rejects, foreign governments can negotiate directly with labs on safety commitments. Expect more lab-government agreements that bypass Washington, such as evaluator access deals with national AI safety institutes.

Part two: middle powers gain influence. With the US out and China cautious, coalitions like the Nordic-led call, the UK G20 effort and Singapore's testing work become the places where rules get drafted. Standards drafted there tend to become de facto global requirements, the same way GDPR did for privacy.

Part three: "superintelligence" loses precision exactly when it is needed. Bills in Congress, EU systemic-risk rules and any future international threshold all need a clear label for the most capable systems. Expect drafters to shift to more technical terms such as compute thresholds, capability evaluations and "frontier model," and to avoid "superintelligence" entirely because the US government has redefined it.

Part four: China gets a free positioning win. By politely declining the new term, Beijing aligns itself with the UN vocabulary and the majority of states, which strengthens its pitch for a UN-centered AI body with a strong Global South role.

Checklist for Global Teams

To compare model costs across providers while the rules settle, use the LLM API Pricing Tracker. For the domestic story and the full text of the order, see our Super Intelligence executive order explainer, and for the research meaning of the term, our look at Sam Altman's superintelligence timeline.

What To Watch Next

• Whether any country beyond the US adopts "super intelligence" in official documents (none had as of September 30)
• More endorsements of the Call for Control of Frontier AI Models, and whether Japan, South Korea or India join
• UN follow-up after France's Security Council presidency ends on September 30
• The late-November US-China AI dialogue and the language used in any joint text
• The UK's 2027 G20 AI standards agenda

Key Takeaways

• The US rename has no legal force outside US federal agencies; the EU, China, UK, UN and standards bodies still say AI
• The Call for Control of Frontier AI Models (launched Sept 21 by Finland and Norway) had 28 country endorsements plus the European Commission by Sept 24; the US and China did not sign
• It asks for mandatory pre-deployment testing, shared incident reporting and a possible international institution
• China "respects" the US term but did not adopt it, a low-cost positioning win for Beijing
• UK plans global AI standards under its 2027 G20 presidency
• At the Sept 23 Security Council, Altman, Amodei and Delangue asked for the global standards the White House rejected
• Redefining "superintelligence" as all AI weakens a key threshold term in treaty and bill drafting
• For global teams: keep standard AI terminology everywhere except US federal paperwork, and add a glossary line

Sources

• Government of Norway, Government of the Netherlands and Portugal: text and signatories of "A Call for Control of Frontier AI Models" (Sept 21 to 25, 2026)
• The Straits Times: Singapore and signatory list; UK G20 remarks (Sept 22 to 23, 2026)
• Ministry of Foreign Affairs of China: spokesperson remarks on AI terminology (Sept 26, 2026)
• AFP via Dawn on the China response and the Trump social media claim (Sept 26, 2026)
• BBC, UN News, TheJournal.ie and The National on the Security Council session (Sept 23, 2026)
• The White House: "Inaugurating the Era of Super Intelligence" (Sept 29, 2026)
• Defense One and BBC on the Bostrom definition and expert views