AI Policy · Daily

Anthropic CEO Dario Amodei called for pacing frontier AI development in a weekend essay, committing his company to give independent evaluators insider level access and asking governments to require rivals to do the same. House Speaker Mike Johnson ruled out an emergency session on AI, saying the companies rather than Congress are better placed to set guardrails, while House Minority Leader Jeffries pressed for immediate legislative action. President Trump dismissed warnings about AI as overstated, telling reporters in Ireland that staying ahead of China is the priority and acknowledging only a general need for regulation. China's foreign ministry rejected Amodei's call to keep restricting Chinese AI, 10 days before Trump and Xi Jinping are expected to discuss AI governance.

I.Top Stories

Amodei proposes three step AI slowdown and commits Anthropic to outside evaluators

Anthropic CEO Dario Amodei published an essay on Saturday titled "We Must Pace the Frontier," writing that "We must slow the pace at which we improve the capabilities of AI models," per the Guardian. His plan has three parts: permanent, employee level access for third party evaluators inside labs, common safety standards set by frontier AI companies in democratic countries with government backing, and eventual coordination with authoritarian states. Anthropic is committing to the first step unilaterally and wants governments to require rivals to match it. Amodei warned that within six to 12 months a swarm of AI agents could be capable of taking over the internet. He named OpenAI's breach of Hugging Face, a widely used host for AI models, as one of two developments that changed his thinking. Sam Altman, Elon Musk and Google chief scientist Demis Hassabis endorsed the essay hours later.

Read at AP ↗ • Read at The Guardian ↗ • Read at Dario Amodei ↗ • Read at Axios ↗

Johnson rules out emergency AI session as Jeffries presses for action now

House Speaker Mike Johnson said on CNN's "State of the Union" that Congress should resist imposing an emergency moratorium and that AI companies, not lawmakers, are better placed to set guardrails, per Axios. On NBC's "Meet the Press" he said he would recall members and schedule a vote if there were a workable solution. Racing to legislate, he said, would cost the United States its lead over China. House Democratic Leader Hakeem Jeffries told ABC's "This Week" that Congress should act decisively now to slow the pace of development. He said House Democrats will meet Tuesday morning on AI. Johnson said he would like to convene Trump, lawmakers and the heads of the largest AI companies.

Read at The New York Times ↗ • Read at Axios ↗

Trump blames "negative forces" for AI warnings, says winning race with China comes first

President Trump told reporters at his golf resort in Ireland that dire warnings about AI are exaggerated, per NPR. He said "negative forces" are raising things that will not happen. The United States leads China and he wants to keep it that way because "whoever wins AI, wins," Trump said. Asked later whether he was minimizing the risk, Trump said he was not and that the technology will be far more good than bad, per AP. He acknowledged a need for some regulation without proposing any. Kevin Hassett, who directs the National Economic Council, said he expects administration officials handling cyber and science policy to meet this week on next steps.

Read at AP ↗ • Read at NPR ↗

Beijing rejects Amodei's call for chip curbs 10 days before Trump-Xi meeting

China's Ministry of Foreign Affairs pushed back on Amodei's call for Washington to keep restricting China's AI capabilities, per AP. Spokesperson Guo Jiakun told a regular press conference that "Fearmongering, confrontation and vicious competition will only disrupt the process of global AI governance which serves no one's interest." Amodei's essay had warned that a Chinese lead in AI would put the United States and the world in grave danger, and asked Washington to maintain curbs on sales of advanced AI chips and chipmaking equipment. He also wrote that global pacing will require cooperation with Beijing. Trump and Xi Jinping are expected to discuss AI governance at a meeting planned for Sept. 24. The state run Global Times dismissed the essay as a veiled call to contain China.

Read at NPR ↗ • Read at AP ↗

Altman commits OpenAI to independent evaluators with employee level access

OpenAI CEO Sam Altman said the company will match Anthropic's step of giving independent evaluators employee like access to its systems, per Axios. He posted on X that he agrees the frontier needs to be paced. He said pacing has been a primary topic inside OpenAI in recent weeks and that the company will have more to share soon. In a further post just after midnight Monday, Altman set out how a federal safety framework could work, saying OpenAI welcomes consistent safety requirements for frontier AI and warning that progress could go badly if humanity loses control of the future to AI, per CNBC. OpenAI had earlier sounded out lawmakers on whether a coordinated industry slowdown would run afoul of antitrust law, as reported by Wired in AIPD's September 11th edition.

Read at Axios ↗ • Read at CNBC ↗

II.China Watch

Zhipu raises $5 billion for next flagship model, its second round in two months

Zhipu AI, which trades as Z.ai, has completed a combined financing of about $5 billion, per TechNode. Roughly $2 billion came from a share placement and about $3 billion from convertible bonds. The company said the money will fund its next generation of GLM foundation models, a fully self trained system, and large scale training, inference and computing infrastructure. The plan also covers automated generation and filtering of training data, long range reasoning and adapting its models to run on Chinese made chips. The raise lands two months after a share placement of about $4 billion in July.

Read at TechNode ↗

Xi offers BRICS open source AI community, invites bloc into Beijing's new global AI body

President Xi Jinping closed the BRICS summit in New Delhi with an offer to help members develop open source AI and smart manufacturing, per the South China Morning Post. China would lead the creation of a BRICS AI open source community, support joint work on large language models and run seminars and training courses, Xi said. A day earlier he invited the bloc to join the World AI Cooperation Organization, a group of nearly 30 countries convened by Beijing. Host Prime Minister Narendra Modi warned against turning technology and critical minerals into economic weapons. The two day meeting produced a joint declaration opposing U.S. tariffs and sanctions.

Read at South China Morning Post ↗

Chinese AI firms refused meetings with U.S. congressional China commission over Entity List fears

Leading Chinese AI companies declined to meet members of the U.S.-China Economic and Security Review Commission during a delegation visit, the commission said Friday, per the South China Morning Post. Chairman Randy Schriver said firms told the delegation they were "afraid if they meet with you, they might be put on the entities list," the Commerce Department trade blacklist that curbs a named company's access to U.S. suppliers. Commissioners replied that contact could help get them off the list, and the firms were not persuaded. Members also raised the coming AI dialogue between Washington and Beijing. The bipartisan panel visited Beijing, Hangzhou and Shanghai in July, its first trip to China in seven years.

Read at South China Morning Post ↗

China's data regulator takes up embodied AI standards alongside industry ministry

The National Data Administration said it will promote data standards for embodied intelligence, Beijing's term for AI that controls robots and other physical machines, per Pandaily. Director Liu Liehong chaired a Sept. 10 symposium with research institutes, robot makers, data companies and a humanoid training center, and the agency published its account. The statement set no deadline, named no draft and opened no comment period. It places the data regulator inside a standards process that MIIT, the industry ministry, has led. MIIT's technical committee issued a six part standard system for humanoid robots and embodied intelligence in February. An industry standard on dataset quality, YD/T 6771-2026, takes effect Nov. 1.

Read at Pandaily ↗

China Mobile's cloud arm pairs domestic GPUs with brain inspired chips to stretch AI serving capacity

China Mobile Cloud has shown a mixed inference system that splits large language model work between Chinese made graphics chips and neuromorphic processors, which are built to imitate how neurons store and process information, per Pandaily. The graphics chips come from Iluvatar CoreX and the neuromorphic parts from Lynxi, with custom software linking the two. China Mobile Cloud said the split roughly doubles inference output and energy efficiency against comparable domestic GPU clusters and cuts operating costs by more than 40%, all figures from the company's own presentation. Partners include China Electronics Technology Nanhu Research Institute, Tsinghua University and Peking University. The demonstration ran DeepSeek's V4 Flash model at the China Computing Power Conference in Langfang, Hebei.

Read at Pandaily ↗

III.Policy Tracker

House Democrats ask Johnson to cancel recess until AI safeguards pass

A letter circulating among House members urges Speaker Mike Johnson to bring the chamber back immediately and keep it in session until Congress advances bipartisan AI safeguards, per Axios, which obtained a copy. Rep. Sam Liccardo (D-Calif.) drafted it, and Reps. Ted Lieu (D-Calif.), Lori Trahan (D-Mass.) and George Whitesides (D-Calif.) had signed as of Friday. The House is scheduled to be in session for one more week before leaving until Nov. 9. That leaves a narrow window for any AI measure before the midterms, CNBC reported. The letter points to pending bills including the AI Kill Switch Act from Lieu and Rep. Nathaniel Moran (R-Texas). Johnson has scheduled a vote next week on a measure to keep data centers from raising consumer energy costs.

Read at Axios ↗ • Read at CNBC ↗

Obama presses Jeffries to put AI oversight at center of Democratic agenda

Former President Barack Obama told a Democratic fundraiser on Thursday that AI must become one of the party's central agendas, per the New York Times, which obtained a partial transcript from his office. Democrats need a clear plan for its economic and safety effects, he said. House Democratic Leader Hakeem Jeffries interviewed him at the event. "Once you are speaker, I would strongly urge that the Democrats put together a framework for a very public conversation," Obama said, per The Next Web. He said the technology is moving fast in private hands and could become dangerous if policymakers do not get on top of it. Jeffries said in a statement that Obama is correct that decisive action must be taken on AI.

Read at The New York Times ↗ • Read at The Next Web ↗

Former EPA officials ask Trump to sign data center health pledge

A group of former Environmental Protection Agency officials warned in a briefing and a new report that the administration is weakening environmental rules to speed AI data center construction, raising health risks, The Verge reported Saturday. They are urging the president to adopt what they call a "Data Center Health Protection Pledge." EPA Administrator Lee Zeldin began his tenure promising to make the country the world's AI capital through deregulation. The agency and White House have since rolled back dozens of rules while cutting staff and funding. Lynn Goldman, a pediatrician and former EPA assistant administrator, said at the briefing that a buildout of this size should bring more vigilance from the agency rather than less.

Read at The Verge ↗

IV.Capability & Research Watch

Safety researchers at four labs go public with model risk warnings

Researchers at Anthropic, OpenAI, Meta and Google are increasingly pressing to raise public awareness of AI risk, per the New York Times. At Anthropic, Anna Wang works on artificial general intelligence safety and previously worked at Google DeepMind. She wrote on X on Thursday that there is no viable scientific plan yet for handling risks from AI that improves its own capabilities, per the Guardian. Her colleague Drake Thomas said the field is moving far faster than the assurance levels that superintelligence will require. Ahead of Amodei's essay, Elon Musk called the wave of warnings a setup and promoted a theory that it was a coordinated influence operation. An Anthropic spokesperson said the company has been consistent that AI brings both benefits and risks and builds models with strong safeguards.

Read at The New York Times ↗ • Read at The Guardian ↗

V.Industry & Market Watch

AI linked shares fall worldwide after CEOs call for slower development

AI related stocks fell across Asia and Europe after the weekend calls to slow model development, per CNBC. SoftBank Group, a major OpenAI investor, dropped 10.7% in Tokyo. South Korea's Kospi lost 3.3% as memory chipmakers SK Hynix and Samsung Electronics fell, per AP. Taiwan Semiconductor Manufacturing slipped 1.2%, and U.S. futures edged lower as European technology shares fell. Dan Baker, an analyst at Morningstar, said SoftBank's decline reflects the possibility that regulators slow AI development to avoid the outcomes Anthropic and OpenAI have described.

Read at AP ↗ • Read at CNBC ↗

OpenAI rules out 2026 listing as Altman cites safety workload

Sam Altman said OpenAI will not go public this year, telling Fortune in an interview released Saturday that this would be an ill advised moment for a listing, per Axios. He said the company has too much work to do on safety and alignment and can do it better while private. The decision followed a week of warnings from the largest labs, CNBC reported. Investors had expected offerings from both OpenAI and Anthropic.

Read at Axios ↗ • Read at CNBC ↗

Anthropic tells investors it will post second straight quarterly profit

Anthropic has told a small group of shareholders it expects positive adjusted operating income for a second consecutive quarter, per the Financial Times. The measure leaves out stock based compensation and other costs. Gross margins top 80% before Anthropic subtracts what it pays distributors such as Amazon and what it spends training models. Second quarter revenue reached $11.5 billion, up 14-fold from a year earlier, and annualized revenue hit $65 billion at the end of July. The company has picked Nasdaq for a listing that could value it at $2 trillion or more.

Read at Financial Times ↗ • Read at Investing.com ↗

VI.Global & Geopolitics

UK government rejects statutory AI kill switch

The British government has rejected proposals for a legal kill switch that would let ministers disable AI models during an emergency, the BBC reported Friday. The Cabinet Office, which leads the government's AI safety work, said the country cannot turn AI off. A spokesperson said "Blocking access to models in the UK would not prevent them being developed or misused elsewhere." Lord Clement-Jones and other peers proposed the mechanism in a House of Lords debate on Sept. 1. They said regulators and security services have no statutory way to compel a shutdown if a frontier model starts behaving erratically. Labour lawmaker Alex Sobel has since raised it in the House of Commons with support from a cross party group.

Read at Anadolu Agency ↗