01
US · China · AI Diplomacy
Developments 9/24–26
US and China Launch "Superintelligence Dialogue" Two Days After the Summit — but the "AI Red Telephone" Comes Without Rules
Differences over Taiwan and Iran remained unresolved at the White House summit, but the AI-incident hotline announced two days later was the week's most concrete achievement.
The Summit — A Grand Welcome, Unresolved Differences
President Trump gave President Xi and his wife an elaborate White House welcome on 9/24. Al Jazeera's four takeaways: ① both sides stressed stable relations — Xi warned of a "Thucydides Trap" and called for keeping "competition within manageable bounds"; ② the Iran war remains unresolved, with China reluctant to back US calls to isolate Iran; ③ reaction in the US Congress was mixed — criticism of "excessive hospitality" alongside Senator Rand Paul's support for "more talking, less war"; ④ Taiwan remains a point of disagreement — Xi urged opposition to "Taiwan independence" while the US maintained strategic ambiguity.
Korea Angle
On the day of the summit (9/24), the KOSPI closed up 0.90% at 7,080.92, Samsung Electronics rose 3.28% to ₩285,500, and SK hynix gained 1.20%. Korean markets had been watching the summit as a swing factor covering "an extended tariff truce, AI regulation, Taiwan and the Iran war" — the gain came as a rebound from the prior week's (9/14) AI-pacing-driven sell-off.
Two Days Later, a Concrete Outcome — the "Superintelligence Dialogue"
On 9/25–26, the White House announced that the US and China had agreed to launch a "US-China Superintelligence (SI) Dialogue." The channel, meant to discuss AI's risks and benefits, includes a national-security-level AI-incident notification channel comparable to the Cold War-era US-Soviet nuclear hotline, with the next round set for Shenzhen this November. Treasury Secretary Bessent led the negotiations. The White House said both sides agreed to refer to artificial intelligence as "superintelligence (SI)" under this agreement — and even as President Trump maintained that "beating China in the AI race is the single most important thing," the agreement effectively acknowledged AI as a serious competitive risk that needs managing. CNBC reported that China separately confirmed a $30 billion tariff cut and an agreement to keep talking.
What's Missing — a Hotline Without Rules
WION assessed the deal as "symbolic, not yet a safeguard" — there is still no definition of what counts as an "incident" that would trigger the channel, no designated officials on either side, and no procedure for private AI labs to report incidents to their governments. "A hotline is only useful if there's a shared understanding of when to pick it up," the outlet wrote. The November Shenzhen round will be the first test of whether real substance gets filled in. On 9/26, China's Foreign Ministry responded to the term "superintelligence" by saying "China values and respects the US side's position and expression. As AI technology keeps evolving, all countries can strengthen exchanges and seek common ground in step with the latest developments" — diplomatic wording that leaves room to maneuver rather than a full embrace of the term.
Importance 5/5 — the first US-China government-to-government AI safety channel, with detailed rules deferred to November's Shenzhen round
02
UN · AI Governance
Developments 9/21–23
Finland and Norway's "AI Oversight Body" Declaration Meets Trump's "Globalist Scheme" — a Day Later, Altman, Amodei and Bengio Make the Opposite Case at the Security Council
Over three days at the UN General Assembly, a call for international oversight from more than 20 governments, a flat rejection from the White House, and an appeal for cooperation from industry's top executives collided in sequence.
9/21 — Finland and Norway Fire the Opening Shot
Finland's President Alexander Stubb and Norway's Prime Minister Jonas Gahr Støre jointly launched a declaration on 9/21 calling for international cooperation to keep frontier AI models under human control. As of reporting on 9/22, 20 countries and the European Commission had signed; by week's end that had grown to 28 countries and 30 heads of state and government — including Germany, Canada, Australia, Singapore, the UAE, Bahrain, Kazakhstan, Kenya and South Africa. The declaration called for establishing an international institution to "convene states, set standards and enable verification when capability thresholds are crossed" — ThePrint likened the model to the International Atomic Energy Agency (IAEA). Stubb: "Humans should be in the driver's seat in this shift. We have choices in how we respond to AI: we can compete, cooperate, or step back. The right choice is to cooperate." Støre: "The development of frontier AI models must be brought under better control... Addressing these risks requires political action." The US and China were absent from the signatory list from the start.
A Day Later, Trump's Flat Rejection
In his address to the 81st UN General Assembly on 9/22, President Trump called international AI oversight efforts a "globalist scheme." He said "we're not going to choke off something that's going to be bigger than the Industrial Revolution," and dismissed warnings of existential AI risk as a "hoax," comparing them to climate-crisis advocacy. He also singled out Dario Amodei by name while suggesting his administration already holds regulatory authority over AI companies. Trump did not name the Finland-Norway declaration directly, but the timing read as an explanation for why the US was absent from its signatory list.
No one person or company or country should be able to use the most powerful AI models to impose their worldview on everyone else.
— Sam Altman, CEO, OpenAI, UN Security Council, 9/23 2026
A Day After That, Altman, Amodei and Bengio Take the Security Council Floor
On 9/23, Sam Altman and Dario Amodei appeared together before the UN Security Council to call for international cooperation and shared safety standards. Amodei: "If managed poorly, I even believe AI could be a risk to humanity as a whole." Canada's Yoshua Bengio also spoke, describing AI risk as "real and imminent" — the same week, the Canadian government announced C$150 million in funding for his AI-safety research organization. Canadian Prime Minister Mark Carney had already signed the Finland-Norway declaration himself — a rare instance of a government declaration, an individual expert's appeal and domestic funding all pointing the same direction within a single week. UN coverage quoted Bengio describing the danger as one "that none can contain alone and that does not respect the borders we defend," and one UN expert described the current state of AI development as "a race that everyone loses." Barely a day after Trump's remarks, the US administration and the leaders of its own AI industry had delivered opposite messages to the world.
Korea Angle — Lee Jae-myung's "Global AI Basic Society," but No Seat on the Declaration
At the same General Assembly, President Lee Jae-myung used his keynote address on 9/22 (local time) to unveil "a global AI basic society" as the first vision of his "K-Initiative." "No nation and no individual should be left behind in AI's benefits," he said, outlining plans to build a domestic "Global AI Hub" involving international organizations, multilateral development banks, companies and academia, and to share Korea's AI-driven transformation in healthcare, social security and labor with the international community. Korea, however, did not sign the Finland-Norway-led declaration of 20 to 28 nations, whose signatories included Singapore, Australia, Canada and the UAE — placing it, alongside Japan, India, the UK and France, among the non-signatories. Between Trump's rejection of oversight, Finland and Norway's "international oversight body," and Altman and Amodei's "appeal for cooperation," Korea offered a fourth framing centered on "universal benefit-sharing" — though it has yet to join the multilateral oversight system itself, a point worth noting alongside its own proposal.
Importance 5/5 — 28 countries, minus the US and China, called for international oversight, while the White House and its own industry sent opposite messages a day apart
03
US · AI Industry
Announced 9/22
"Where Did the Slowdown Go?" — Anthropic and OpenAI Launch New Models Simultaneously, and a Price War, Ten Days After Calling for a Pacing Slowdown
The same companies that rattled the market with a "slowdown" call last week released new models on the very same day — and cut prices.
Same Day, Opposite Direction
On 9/22, Anthropic and OpenAI unveiled new models on the same day — Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol and Luna. OpenAI priced its API below Anthropic's; Opus 5.5 costs roughly 40% less to run than Opus 5, while OpenAI also cut prices about 50% below its previous promotional rates. Anthropic itself said Opus 5.5 was "its first release since Amodei's 'Pace the Frontier' statement" — exactly ten days after the 9/12 pacing proposal.
Expert Take — "A Price War That Eats Its Own Profitability"
One economics commentator said "OpenAI and Anthropic are locked in a price war driving AI prices down, which in turn is eating into their own ability to generate returns." The two companies that had called for slowing down and coordinating ended up in one of the most intense forms of competition — simultaneous model launches and price cuts — continuing the "paradox of pacing rhetoric alongside compute expansion" flagged in last week's briefing.
Importance 3/5 — a case study in the gap between "slowdown" rhetoric and actual competitive behavior
Policy Calendar · Background Context
The items below are not confirmed news from this week (9/21–27), but scheduled events and ongoing matters included for reference.
November
Second round of the US-China "Superintelligence (SI) Dialogue" — expected in Shenzhen, China (Story 1) — whether concrete rules get filled in is the key question
Ongoing
The Finland-Norway-led oversight declaration — signatories grew from 20 countries on 9/22 to 28 by week's end (Story 2); next steps (negotiating a formal institution) remain undecided
Ongoing
The UN General Assembly's 81st high-level week continues — more national addresses and AI-related follow-up remarks expected
Ongoing
The UN's Independent International Scientific Panel on AI — with Bengio among its participants, its preliminary findings were cited again at the Security Council briefing (Story 2)
To confirm
The $30 billion US-China tariff-cut agreement (Story 1) — specific items and implementation timing still need confirmation
This Week In Sum
"Separate Rules, a Shared Channel" — and a Fractured Push for International Oversight at the UN
This week, the US and China set up an AI safety-incident channel just two days after their summit (9/24), establishing a shared baseline of "manageable competition" — yet the channel still has no trigger criteria and no designated officials. At the UN General Assembly the same week, a different fault line opened up: a declaration led by Finland and Norway, backed by 20 to 28 countries, called for an IAEA-style international oversight body, but the US and China were absent from the start, and President Trump dismissed it the very next day as a "globalist scheme." A day after that, Sam Altman, Dario Amodei and Yoshua Bengio made the opposite case for international cooperation before the Security Council — Canada straddled both tracks, having signed the declaration and funded Bengio's work (C$150 million) in the same week. President Lee Jae-myung offered a fourth framing centered on "universal benefit-sharing," even as Korea itself did not sign the Finland-Norway declaration. Anthropic and OpenAI, having called for a slowdown, undercut their own rhetoric by launching new models simultaneously the same week — whether through government channels, multilateral oversight, or industry self-regulation, the gap between declaration and execution was this week's recurring theme.
Key Takeaways
1. US-China AI Relations Started From a "Channel," Not "Rules"
The SI Dialogue matters less for what was agreed than for the fact that an agreement exists at all — even without definitions, designated officials or a reporting framework, both sides publicly confirmed that "a channel exists." Whether it turns from symbol into substance by the November Shenzhen round is the real test, and any AI-related incident or accidental friction in the meantime is still left to a hotline with no rules attached.
2. A New Axis — a "Middle-Power Coalition" — Has Emerged, and Korea Isn't in It Yet
The 28-country declaration led by Finland and Norway is the first time countries outside the US and China rallied around a single demand for "international oversight." Middle powers often compared to Korea — Singapore, Australia, Canada, the UAE — signed on in large numbers, while Korea, Japan, India, the UK and France did not. President Lee Jae-myung's UN address, "a global AI basic society," charts an independent course centered on benefit-sharing rather than oversight — but since Korea has yet to join the multilateral oversight system itself, its position within this week's middle-power coalition remains unsettled.
3. The Split Runs Inside the US Too, Not Just Between Countries
One day after the Trump administration dismissed international AI oversight as a "globalist scheme," Altman and Amodei — representing US industry — delivered the opposite appeal for international cooperation at the UN. The "accelerate / decelerate / self-regulate" three-way split flagged in last week's briefing isn't only playing out between countries — it's recurring between a single country's own government and its industry. Reading AI governance purely at the level of nations misses this kind of fracture.