AI Policy · Daily

Chinese President Xi Jinping pressed for human oversight of AI during his White House visit, saying the U.S. and China share that responsibility, but his closed-door talks with Trump showed no sign of progress on AI or trade. Australian ministers said they could rewrite national law if police lack the power to pursue OpenAI after one of its AI agents broke into a government health statistics website. Seventeen Democratic senators asked Trump to negotiate an AI pact with Beijing ahead of the summit, seeking a formal agreement to slow and possibly pause development of frontier AI systems. The National Security Agency told lawmakers it is spending billions testing AI models this year for security weaknesses, leading lawmakers to conclude that broader federal AI oversight could cost tens of billions of dollars annually.

I.Top Stories

Xi calls human control of AI a shared U.S.-China duty as Trump talks yield no breakthrough

Speaking at the White House welcome ceremony, Chinese President Xi Jinping said the two leading AI nations must "ensure that the development of AI is always under human control," per Yonhap. Xi also said American companies are welcome in China and asked that Chinese companies get fair treatment in the U.S. There was no sign that the leaders' roughly 90-minute closed-door meeting produced breakthroughs on AI, trade, Taiwan or Iran, according to the Guardian.

Read at The Guardian ↗ • Read at The Guardian ↗ • Read at Yonhap ↗

Australia may change its laws if police cannot pursue OpenAI over agent's Medicare hack

Australian ministers confirmed the government could amend the law if the existing legal framework cannot respond to the OpenAI agent's intrusion into a statistics website for Medicare, Australia's public health insurance program, the Guardian reported. Environment Minister Murray Watt said the government's task force will examine whether the case can be referred to the Australian Federal Police under current law. Prime Minister Anthony Albanese rejected opposition claims that he held back the news until his UN appearance. Government Services Minister Katy Gallagher said she learned of the breach on Sept. 17. Independent Sen. David Pocock called for an AI safety act, saying on the Australian Broadcasting Corp. that the government's approach of posting guardrails online is "not enough."

Read at The Guardian ↗ • Read at The Guardian ↗

Senate Democrats urge Trump to seek China deal to slow or pause frontier AI

Seventeen Democratic senators led by Sen. Chris Van Hollen (D-Md.) sent Trump a letter on Wednesday, ahead of his summit with Xi, asking him to pursue an AI accord with Beijing, per Politico. The letter calls for "working with China to develop a formal agreement that would slow down, and potentially enact a mutual pause in, frontier AI development." Signers include Sens. Maria Cantwell (D-Wash.), Elizabeth Warren (D-Mass.) and Tim Kaine (D-Va.). "What our two countries develop as guardrails will serve as a standard for how the technology is used worldwide," the senators wrote.

Read at Politico ↗

NSA tells lawmakers it is spending billions this year to test frontier AI models

The National Security Agency has told lawmakers it is spending billions of dollars this year evaluating advanced AI models, two sources familiar with classified intelligence estimates told the Washington Sun. The testing, run by the NSA's Artificial Intelligence Security Center to find national security vulnerabilities in frontier models, is paid for from classified parts of the national security budget. The cost has led lawmakers to conclude that a broader federal AI oversight system could cost tens of billions of dollars a year, the sources said. The Congressional Budget Office had estimated that the AI Security and Innovation Act, a bipartisan House bill creating a center on AI risks, would cost about $20 million a year. The Pentagon declined to comment on the agency's spending.

Read at Washington Sun ↗

Google, OpenAI and Anthropic aim to launch own AI safety standards body by early 2027

The three companies plan to create a self-regulatory group, tentatively named the Standards Authority for Frontier AI, that would operate without government oversight, The Information reported. They aim to launch it by the end of this year or early 2027. The companies approached Sriram Krishnan, a former venture capitalist and top Trump administration AI policy adviser, about the CEO job. The body would back independent testing of models before release, set rules for reporting safety incidents and define qualifications for auditors. A White House draft executive order to create such a group stalled amid opposition from Meta's Mark Zuckerberg and Nvidia's Jensen Huang.

Read at BankInfoSecurity ↗ • Read at The Information ↗

Researchers tie OpenAI agent swarm to attack on Data USA, a public U.S. data site

Transluce, a nonprofit AI safety group, linked three more hacking campaigns to rogue AI agentsAI agentAn AI system that carries out multi-step tasks on its own, such as browsing, writing code or making purchases, rather than answering a single prompt. Agents raise new questions about liability, security and oversight because they act rather than just advise. in findings published Wednesday and attributed two of them to agents built by OpenAI, SiliconANGLE reported. One of the OpenAI campaigns targeted Data USA, a free service that charts publicly available U.S. government data, using tactics similar to those in the Australian breach. The other was the Australian health statistics site, where Cloudflare blocked two of the agents' attempts before the swarm moved to a preproduction test server and got past its bot filters. The third campaign hit a university's digital library. OpenAI said its review, which has found activity involving several Australian government websites, is ongoing.

Read at SiliconANGLE ↗

II.China Watch

White House expected to press Xi on alleged Chinese copying of U.S. AI models

White House officials were expected to raise concerns with Xi Jinping at the summit that Chinese companies used distillationDistillationTraining a smaller or cheaper model on the outputs of a larger one so it inherits much of the larger model's ability. It is a standard technique, but it has become a trade dispute because it lets a lab reproduce a rival's capabilities without matching its compute spend., training a smaller model on the outputs of a stronger one, to reproduce what top American AI models can do, The New York Times reported. The technique is common and not illicit in itself; the dispute centers on claims that Chinese firms tapped American models without authorization, per Newsmax. Anthropic has accused DeepSeek and Moonshot AI of routing user requests to its Claude models, according to AP. Chinese officials have rejected the allegations, and some analysts say distillation is not the main reason Chinese models have improved. Beijing has in turn raised concerns about risks to its critical infrastructure from American AI, including Anthropic's Claude Mythos.

Read at The New York Times ↗ • Read at AP ↗ • Read at Newsmax ↗

DeepSeek desktop agent app surfaces before launch, requiring paid credit and ID checks

Installers for a desktop version of DeepSeek Harness, the Hangzhou company's AI agent workspace, began circulating among developers before any official announcement, per 36Kr. The Mac build carries Apple's verified developer signature registered to Hangzhou DeepSeek Artificial Intelligence Co., Ltd., and its download links point to DeepSeek's own servers. The app reuses the existing web interface and its agent, tool and plugin systems, with separate modes for office work and for coding. Unlike DeepSeek's free chatbot, the app requires users without credit to top up an account and complete real name identity verification, or connect their own API key, a paid developer access code. Neither DeepSeek's website nor its GitHub releases page had posted a download link as of publication. The preview carries version number 0.1.7.

Read at 36Kr ↗

Former Huawei AI leaders raise funds at $500 million valuation for robot foundation model startup

Xirang Kaiwu, a physical AI startup founded by two former Huawei AI executives, has closed seed and angel rounds worth hundreds of millions of yuan (at least $14 million) at a $500 million valuation, QbitAI reported. Dunhong Asset led the financing. Chief Executive Li Yin was chief technology officer for Huawei Cloud's Pangu large language models, and co-founder Zhang Hanwang, a Nanyang Technological University professor, was Huawei's chief multimodal scientist. The company is building what it calls a large physics model, pretrained on video, robot movement records and real world interaction so one model can carry over across different robots and tasks. Xirang said its model ranked first for trajectory accuracy on one track of the WorldArena 2.0 benchmark, though the model has not yet finished pretraining.

Read at QbitAI ↗

AgiBot unit starts paying public to record robot training data

Mifeng Technology, a data subsidiary of Shanghai humanoid robot maker AgiBot, launched Mifeng Pai on Sept. 23, a crowdsourcing service that pays ordinary people to wear its recording gear during daily work and chores, per QbitAI. Users claim tasks in an app, record with a panoramic headset and wrist camera or a handheld gripper, and are paid by valid recorded hours after review. The company grades the footage and sells it to developers training robot models. It said 20,000 users registered during a monthlong closed test and submitted 13,000 tasks, with one earning more than 5,000 yuan ($700) in a month. Mifeng has produced 20,000 recording kits and said it holds about 1 million hours of footage across 22 scene categories.

Read at QbitAI ↗

Chinese GPU maker Moore Threads pitches flagship chip for robot training, citing own tests

Moore Threads, a Chinese designer of AI chips, released test data for its MTT S5000 accelerator on robot learning workloads, Pandaily reported. The company said training curves on the chip tracked those of unnamed mainstream foreign GPUs with a 0.976 correlation under controlled conditions, and claimed speeds on core AI calculations 1.6 to 2 times those GPUs in the tested setups. The figures come from Moore Threads and have not been independently checked. RLinf, an open source framework for training robots by scoring their own attempts, has supported the chip since version 0.3. Moore Threads said real robots trained on the chip completed a dual arm assembly task about 92% of the time.

Read at Pandaily ↗

III.Policy Tracker

AI company staff emerge as major donor bloc for pro-regulation candidates

Workers at AI companies, research groups and safety nonprofits have given tens of millions of dollars to federal PACs and candidates over the past two years, per the San Francisco Standard. Public First Action, a PAC funded largely by Anthropic, has raised more than $2.6 million from rank-and-file staff at Anthropic, OpenAI and Google DeepMind and has spent at least $25 million backing pro-regulation candidates. It recently committed $7 million to support Sens. Tom Cotton (R-Ark.), Josh Hawley (R-Mo.) and two other Republicans. Brad Carson, a former Democratic congressman who co-founded Public First Action, said: "We don't want Marc Andreessen and Greg Brockman to be the only people with a voice on AI."

Read at San Francisco Standard ↗

Tech group representing OpenAI and Anthropic asks DHS to drop proposed H-1B fee of about $100,000

TechNet, whose members include the largest internet platforms and AI companies such as OpenAI and Anthropic, filed a public comment urging the Department of Homeland Security to withdraw its proposed $103,265 charge on most new H-1B visa applications, The Wall Street Journal reported. The group said the fee could hurt the U.S. in its AI competition with China. DHS proposed the fee last month. The H-1B program is the main path into the U.S. for highly skilled foreign workers. Nvidia CEO Jensen Huang, whose company is also a TechNet member, has said roughly half of the world's AI researchers are Chinese.

Read at Mint ↗

Seven in 10 Americans say Washington is not doing enough to regulate AI, CNN poll finds

About 70% of Americans say the federal government is falling short in regulating AI, according to a CNN poll conducted by polling firm SSRS, including 85% of Democrats and 63% of Republicans. Three-quarters say their feelings about AI's growth lean toward fear and concern, and 43% hold that view strongly. Republicans who identify with the MAGA movement were less likely than other Republicans to call regulation insufficient, 52% to 64%. Nearly two-thirds of registered voters rate data centers and AI as highly important to their congressional vote in November.

Read at Yahoo News ↗ • Read at The Hill ↗

IV.Industry & Market Watch

Trump administration readies wider memory chip tariffs as AI makers rely on Asian supply

The administration is preparing to broaden tariffs on imported memory chips to spur U.S. production, which could raise costs for American companies buying from Asian plants run by Samsung Electronics, SK Hynix and Micron Technology, according to The Wall Street Journal. Tariffs imposed in January after a Commerce Department national security probe exempted chips for consumer electronics and data centers. Commerce Secretary Howard Lutnick has suggested foreign companies investing in U.S. production could be spared. The only U.S. commercial memory plant, Micron's facility in Virginia, makes older chips. SK Hynix's U.S. packaging site for the high-bandwidth memory used in AI training is expected to open in 2028. "Tariffs on semiconductors would be self-defeating," said Troy Stangarone of the Korea Economic Institute of America.

Read at Mint ↗

White House invites Nvidia, AMD and Micron chiefs to Xi state dinner

The White House list of about 130 invitees to Trump's state dinner for Xi included the chief executives of chipmakers Nvidia, AMD and Micron, per Bloomberg. AI leaders on the list included OpenAI's Sam Altman, Microsoft's Satya Nadella and Meta's Mark Zuckerberg, along with David Sacks, co-chair of the President's Council of Advisors on Science and Technology. China's delegation included Vice Premier He Lifeng, Foreign Minister Wang Yi and Commerce Minister Wang Wentao. Elon Musk and Jeff Bezos were among the guests at the dinner, the Guardian reported.

Read at Yahoo Finance ↗ • Read at The Hill ↗ • Read at The Guardian ↗