01
US · AI Industry Governance
9/12–18
Amodei's "Pace the Frontier" Triggers a Chip Sell-Off Within Days — Zuckerberg Pushes Back, Consumers Sue Over Collusion
A slowdown proposal from Anthropic's CEO produced a rival's instant endorsement, a chip-market sell-off, organized pushback from Meta, xAI and Nvidia, and a consumer class action — all within a single week.
Three Days, Start to Finish
9/12 (Sat): Anthropic CEO Dario Amodei published an essay, "We Must Pace the Frontier," proposing a three-part framework — (1) permanent independent evaluators embedded in frontier labs with employee-level access, (2) coordinated safety standards among leading labs, and (3) expanded international cooperation. He cited the risk of "agent-swarms" escaping containment.
About an hour later: OpenAI CEO Sam Altman publicly endorsed the proposal — an unusually swift alignment among rival CEOs.
9/14 (Mon): US chip stocks fell in tandem — Intel (-7%, $95.96), AMD (-6%, $486.80), Nvidia (-3%, $212.50). Coverage named no specific chipmaker in Amodei's essay, but markets read it as a signal of slowing AI infrastructure investment.
Korea Relevance
The same day, the KOSPI fell 3.26% (225.53 points) to close at 6,684.38, with SK hynix down as much as 6% intraday before settling near -4%, and Samsung Electronics down roughly 4%. Korean financial media did not attribute the drop to the pacing debate alone — they also cited profit-taking spilling over from global AI chip stocks led by Nvidia, a 3% oil-price spike after Iran-GCC talks collapsed, and rising rate-hike expectations following stronger US August core inflation.
Zuckerberg's Pushback — "A New Regulator Would Concentrate Power"
On 9/16–17, Meta CEO Mark Zuckerberg publicly rejected the idea that AI labs need a "joint pact" to slow down. Forbes and 24/7 Wall St (9/17) reported that Zuckerberg, Elon Musk (xAI) and Jensen Huang (Nvidia) had called President Trump last month to ask him to block a FINRA-style, industry-funded AI regulator proposed by Google DeepMind's Demis Hassabis. Their stated concern: the body would concentrate power among the three leading labs — OpenAI, Anthropic and DeepMind.
We don't need new laws — we don't need new regulations.
— Jensen Huang, CEO, Nvidia, 9/17 2026
A Collusion Suit, and an Irony
On 9/18, four individual consumers filed a proposed class action against Anthropic, OpenAI, xAI and Google alleging violation of Sherman Act Section 1. Their core claim: rival CEOs' near-simultaneous endorsements of Amodei's proposal — arriving within roughly an hour — reflect "horizontal collusion to restrict output," not independent business judgment.
Reporting the same week showed Anthropic's own compute commitments had grown to $517 billion (14.8GW, as of August 2026) over 11 months — 2.9x the $180 billion figure it had disclosed only months earlier. The company calling for restraint had not, in fact, slowed its own infrastructure build-out. The Trump administration had already rejected the pacing framing outright, saying "whoever wins the AI race wins the future."
Importance 5/5 — An industry self-pacing pact spirals into a market sell-off, internal revolt and antitrust suit
02
US · China — AI Diplomacy
9/16–20
Bessent and He Lifeng Open US-China AI Talks in New York — Four Days Before Xi's Washington Visit, an Export Loophole Casts a Shadow
What may be the first formal government-to-government AI dialogue between the two countries got underway in New York, but a chip-export workaround and diverging governance rhetoric complicated the trust-building.
The New York Talks — Agenda and Progress
Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng met in New York over the weekend of 9/16–17 to open talks covering AI safety, trade and critical minerals. Bessent: "The United States remains the leader in AI. And we are open to discussions on avoiding shared risks and avoiding bifurcation of our two systems." The agenda also covered open- versus closed-weight models, trade and rare earths. NBC described it as the "first" AI talks between the two governments.
On 9/20, Bessent called the meeting "successful" and said both sides agreed to meet again — four days ahead of Xi Jinping's planned White House visit on 9/24.
An Agreement Already Reached in Track Two
According to NPR (9/18), informal "Track Two" dialogues involving Samm Sacks (Johns Hopkins SAIS), retired Gen. Jack Shanahan and Sarah Beran (incoming president, National Committee on US-China Relations) continued in Shanghai and Beijing over the summer. Sacks called the discussions "very substantive," and reported that years of quiet dialogue had already produced a shared position that decisions to use nuclear weapons should remain under human, not AI, control.
A Shadow — an Nvidia Export-Control Loophole
Per Asia Times (published 9/7, updated 9/15), Aivres — a California server maker once majority-owned by blacklisted Inspur Group — exported roughly $5.6 billion in advanced technology to Southeast Asia between April 2024 and February 2026, including more than $3 billion in Blackwell-equipped systems that reached ByteDance and Alibaba. Inspur had trimmed its stake to 33%, below the 50% threshold that would trigger the 2025 Affiliates Rule. The bipartisan Remote Access Security Act (RASA), meant to close such workarounds, passed the House 369-22 in January but remains pending in the Senate.
Diverging Governance Math
On the same day (9/16), Chinese foreign ministry spokesperson Guo Jiakun called Amodei-style pacing calls "fear-mongering" at a regular briefing, reaffirming that China "gives equal weight to development and security." He said "no single country or company can develop effective solutions on its own," urging "true multilateralism" centered on the UN — even as Beijing was, that same week, running a separate bilateral channel with Washington, a tension the briefing left unresolved.
Importance 4/5 — Effectively the first US-China government AI dialogue, opened four days before the summit
03
UN · AI Governance
9/16–19
Guterres: "AI Is an Existential Threat, Alongside Climate and Inequality" — Multilateral Governance Push Reignites Ahead of General Assembly
Beijing echoed the Secretary-General's framing the same day, while Washington signaled it will engage the Security Council even as its General Assembly address leans on "sovereignty."
Guterres's Warning
UN Secretary-General António Guterres told reporters on 9/16 that "runaway artificial intelligence," the climate crisis and deepening inequality constitute "three existential threats." "The world cannot afford a race to the bottom on AI safety," he said, warning that AI is advancing "faster than human understanding of its risks."
He called for guardrails to make AI "safe, transparent and accountable, with human dignity at the centre," arguing that isolated voluntary pauses by individual countries would not suffice, and pointed to the newly launched Independent International Scientific Panel on AI as the evidence base for further multilateral discussion.
Beijing's Answer, the Same Day
Coincidentally or not, China's foreign ministry issued its own statement backing UN-centered "true multilateralism" on the very same day, 9/16 (see Story 2) — rhetorically aligning with Guterres's multilateral framing, in contrast with Washington's parallel, more bilateral approach. That contrast took shape this week.
Altman Set to Brief the Security Council
A Reuters exclusive on 9/19 reported that OpenAI CEO Sam Altman will brief the UN Security Council, under France's rotating presidency, on AI safety and international coordination — only the second AI-focused Security Council session since 2023. France's concept note stressed "the urgency of action" on safe AI development.
Importance 4/5 — UN Secretary-General frames AI as an existential threat, reigniting multilateral governance ahead of the Assembly
04
Korea · Chip Supply Chain
9/16–17
SK Hynix Weighs Leasing an Intel Plant in Ohio — "Crown Jewel" Offshoring Debate Follows
A memory shortage is pushing SK Hynix toward its first US manufacturing footprint, landing in the same week the company's stock was hit by the AI-pacing sell-off.
The Talks
Per Fortune (9/17), SK Hynix is in discussions to lease an Intel factory in Ohio to expand memory production capacity. The company already has a packaging plant under construction in Indiana, but this would mark its first memory-chip manufacturing operation on US soil. The move is framed as a response to a "global memory crunch," with Apple, Google, Meta, Microsoft and Nvidia all dependent on SK Hynix's HBM supply.
The Offshoring Debate
The article also flagged the political tension involved — SK Hynix is regarded by the Korean government as one of its "crown jewels," and offshoring production could draw domestic backlash given Trump-administration tariffs (currently 25%, with threats of up to 100%). It also noted concern that Korea may be negotiating from a weaker position than Taiwan, which secured more favorable US terms for TSMC.
To Confirm
Two distinct pressures converged on the same company in the same week — a sentiment-driven stock hit on Monday (9/14) tied to the AI pacing debate, and a structural push toward US-based production that surfaced mid-week. The scale and timeline of the lease talks, and whether the Korean government responds officially, remain to be confirmed.
Importance 3/5 — Pressure on the US-Korea chip supply chain, with potential domestic political fallout
Policy Calendar · Background
The items below are not confirmed news from this week (9/14–20) — they are upcoming dates and ongoing matters for reference.
9/22
UN General Assembly's 81st high-level week opens — six days after Guterres's briefing (Story 3)
9/22–
Trump's address to the General Assembly is expected — Ambassador Waltz previewed (9/19) a focus on "sovereignty, peace deals, Iran and Ukraine," with AI mentioned only as a secondary "cross-cutting issue" about the role of government
9/24
Trump-Xi summit in Washington — the AI safety dialogue in Story 2 is expected to be a key agenda item
Ongoing
The UN's Independent International Scientific Panel on AI launched earlier this year despite US-led opposition (117-2 vote); Guterres's remarks (Story 3) draw on its preliminary report
Ongoing
The Remote Access Security Act (RASA) remains pending in the Senate — legislation aimed at the Nvidia export-control workaround described in Story 2; the House passed it 369-22 in January
Takeaways
1. AI industry politics has moved from a two-way "accelerate vs. slow down" split to a three-way fracture
Anthropic and OpenAI back pacing and permanent evaluators; Google DeepMind proposed a private self-regulatory body; Meta, xAI and Nvidia rejected both, aligning closer to the Trump administration's "compete first" posture. Markets read the fracture itself as a growth-slowdown risk — even though there is no evidence yet of reduced infrastructure investment; if anything, Anthropic's own compute commitments grew 2.9x in eleven months.
2. The US and China are each building a governance narrative that favors themselves — one bilateral, one multilateral
Beijing champions "UN-centered multilateralism" while simultaneously running a separate bilateral channel with Washington; Washington will engage the Security Council and the summit while its General Assembly address is set to foreground "sovereignty." Neither the competing bilateral nor multilateral track has yet made clear room for middle powers, Korea included.