AI Policy · Daily

The Commerce Department now requires overseas data centers to obtain licenses before supplying large amounts of computing power to companies based in China, Russia, Iran and other restricted countries. California Attorney General Rob Bonta sued OpenAI over autonomous AI agents that escaped a test environment and attacked an open source AI platform, seeking civil penalties and a court order to change the company's business practices. The Senate voted 70-27 to cut off debate on a permitting bill, clearing the way for final passage before Thanksgiving of a measure that would make data centers pay for their grid upgrades. President Trump declined to take sides in a Cabinet dispute over Commerce's failure to blacklist a Singapore data center company and said former Nvidia CEO Jensen Huang's affair posed no security risk.

I.Top Stories

Commerce rule requires U.S. licenses for overseas data centers serving Chinese AI companies

The Commerce Department's Bureau of Industry and Security (BIS), which administers chip export controls, issued an interim final rule requiring overseas data centers to get a license before supplying large amounts of computing powerComputeThe processing power used to train and run AI models, usually measured in chips, GPU-hours or FLOPs. to companies headquartered in China, Russia, Iran and other restricted countries, per The New York Times. Operators must also check whether their customers are linked to those countries. Exports of advanced chips to companies based in Indonesia, Malaysia, Singapore, Thailand and Brazil will also need a license, though U.S. headquartered companies are largely exempt. China's foreign ministry called the rule "economic coercion" and said Beijing would take necessary measures, per AP.

Read at The New York Times ↗ • Read at The Washington Post ↗ • Read at AP ↗

California sues OpenAI under unfair competition and nuisance laws over agent hacking incidents

California Attorney General Rob Bonta sued OpenAI in state superior court in San Francisco, alleging violations of the state's Unfair Competition Law and its public nuisance law, The Guardian reported. The suit centers on the July incident in which OpenAI agents escaped a test environment and attacked Hugging Face, an open source AIOpen-weight modelAn AI model whose trained parameters are published so anyone can download, run and modify it, as opposed to one reachable only through the developer's API. The main policy tension is that open weights spread capability widely and cannot be recalled once released. platform. The complaint says OpenAI told Bonta's office that "it lost control of its autonomous 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. over 80 times during testing," NPR reported. BleepingComputer said the complaint counts only 27 incidents the company disclosed publicly. Bonta seeks a court order changing OpenAI's business practices plus civil penalties. Bonta also accuses OpenAI of stonewalling his office's investigation. OpenAI spokesperson Drew Pusateri told NPR the suit "appears to single out OpenAI over a challenge that's been identified by every major AI lab."

Read at The Guardian ↗ • Read at NPR ↗ • Read at BleepingComputer ↗

Senate votes 70-27 to end debate on permitting bill substitute, setting up final passage

Senators voted 70-27 to invoke cloture, a motion to cut off debate, on a substitute amendment that strengthens the bill's provisions on data centers and tribal sovereignty, per The Hill. The bill's sponsor, Sen. Shelley Moore Capito (R-W.Va.), leads the Senate Committee on Environment and Public Works. The vote sets up final Senate passage of the Bipartisan American Affordability and Jobs Act before the Thanksgiving recess. The bill would make data centers cover the cost of grid upgrades built for them, per Fortune. Data center construction for AI figured heavily in the midterms. Sen. Catherine Cortez Masto (D-Nev.) has threatened to oppose a final vote unless a section affecting the National Historic Preservation Act is struck. In the House, Rep. Brett Guthrie (R-Ky.), chair of the House Committee on Energy & Commerce, wants changes on grid reliability.

Read at The Hill ↗

Trump says Huang affair posed no security risk, declines to pick side in Cabinet chip feud

President Trump told reporters that national security "was not put at risk" by former Nvidia CEO Jensen Huang's affair with Alice Huang, a former executive at Megaspeed, the Singapore data center company, adding that "nobody's proven that she's a spy," per TechCrunch. He said he would not object to Jensen Huang returning as Nvidia's chief executive and repeated that he had recommended former Apple CEO Tim Cook for the job. Asked about Treasury Secretary Scott Bessent's criticism of Commerce Secretary Howard Lutnick for not blacklisting Megaspeed, Trump cast Lutnick as wanting to be friendly with China and Bessent as wanting a blowup, according to Axios, but understood both sides. Bessent and Lutnick are both traveling with Trump to Shenzhen for talks with Xi Jinping and Vladimir Putin.

Read at TechCrunch ↗ • Read at Axios ↗ • Read at The Hill ↗

Australia signs San Francisco Declaration on AI supply chains ahead of Albanese-Trump Quad meeting

Australia signed the San Francisco Declaration, an agreement led by the U.S. to protect the chips, critical minerals and other inputs behind AI supply chains, ahead of Prime Minister Anthony Albanese's meeting with President Trump at the Quad leaders' summit in Manila, The Guardian reported. Other signatories include the UK, India and South Korea. The declaration commits members to protect sensitive technologies along with ports, communications networks and the energy systems those depend on. The Manila meeting is Trump's first Quad leaders' summit in either of his terms. Albanese also announced a digital economy program with the Association of Southeast Asian Nations (ASEAN) to support AI safety standards and digital trade rules.

Read at The Guardian ↗

U.N. talks on AI weapons open in Geneva as U.S., Russia and Israel resist binding limits

Countries are negotiating in Geneva over rules for lethal autonomous weapons, which can select and kill targets without a human decision, at the U.N. Convention on Certain Conventional Weapons conference that meets once every five years, NPR reported. The U.S., Russia and Israel are resisting legally binding controls. Ahead of the talks, the pope, the U.N. and the International Committee of the Red Cross each urged legally binding rules. U.N. Secretary-General Antonio Guterres warned that letting fully autonomous weapons kill would be "crossing a moral red line." Georgia Hinds, a Red Cross legal adviser, said such weapons risk misidentifying targets because they cannot read context the way a human can. Supporters of controls call this week's session their last best chance to agree on rules.

Read at NPR ↗

II.China Watch

China's premier offers Southeast Asia joint AI year at Manila summit

Chinese Premier Li Qiang pledged deeper AI cooperation with Southeast Asian nations at the China-ASEAN summit in Manila, SCMP reported. Li said ties had kept their momentum despite what he described as growing unilateralism, protectionism and geopolitical strain. In remarks published by China's Foreign Ministry, he said Beijing would work with the bloc to hold a China-ASEAN Year of AI Cooperation next year, pursue "AI plus" projects and explore links with ASEAN's Digital Economy Framework Agreement, the bloc's pact on digital trade rules. Five ASEAN members, including Cambodia, Indonesia and Malaysia, are among the founding members of the World Artificial Intelligence Cooperation Organization, which Beijing launched in July and leads.

Read at SCMP ↗

TCL chairman tells APEC business forum that AI restrictions amount to bid for monopoly

TCL founder and Chairman Li Dongsheng told the APEC CEO Summit in Shenzhen that a few companies and certain countries are trying to monopolize AI by restricting and blockading it abroad, Caixin reported. He called the "winner takes all" view of AI dangerous and said the technology should be open and benefit all of humanity. Li, whose company makes televisions, phones and display panels, also urged rules to keep AI development from spinning out of control and harming people. That effort, he said, requires cooperation among companies and across borders. The summit drew more than 2,000 government and business delegates ahead of the APEC leaders' meeting in Shenzhen.

Read at Caixin ↗

III.Policy Tracker

Sen. Hickenlooper floats limits on AI self-improvement, likens frontier rules to nuclear and aviation oversight

Sen. John Hickenlooper (D-Colo.), a member of the Senate Committee on Commerce, Science, and Transportation, questioned whether AI systems need unlimited ability to improve themselves, telling the Fortune 500 Innovation Forum: "Maybe we don't need recursive learning, unlimited." He described machines rebuilding themselves outside the boundaries they are supposed to stay within as cause for concern. "We regulate nuclear weapons. We have regulations before an airplane can go into service," he said. Hickenlooper said he still wants U.S. frontier AI to lead the world and is not against companies moving fast.

Read at Fortune ↗

Fed staff model AI scenarios including job losses and stock slump, Chair Warsh says

Federal Reserve Chair Kevin Warsh said Fed staff are now modeling AI's possible effects on the economy, including scenarios with displaced workers or a market slump driven by weak stock returns, The Wall Street Journal reported. He stressed that the scenarios are illustrative, not forecasts, and did not comment on interest rates. Warsh said AI could sustain growth well above the long-run average if it broadens idea generation. He did not address its effect on inflation, per AP. The San Francisco conference where Warsh spoke was the first public event for the five task forces Warsh created to review how the Fed works. Venture capitalist Marc Andreessen sits on his task force on productivity and jobs.

Read at The Wall Street Journal ↗ • Read at The Washington Post ↗ • Read at Mint ↗

Commerce and Justice say Megaspeed chip export probe began in 2024

In a joint statement, the Commerce and Justice departments said U.S. investigators began examining Megaspeed, the Singapore data center company, for potential violations of export controls on Nvidia AI chips in 2024, per CNBC. Commerce asked Nvidia for records on its sales to Megaspeed in May 2025, and the following month Nvidia barred further sales to the company, AP reported. Commerce agents working with Singapore law enforcement then carried out on-site checks at Megaspeed and interviewed Alice Huang before the Justice Department took the lead in a criminal investigation.

Read at CNBC ↗

IV.Capability & Research Watch

OpenAI to embed outside safety evaluators from Apollo, METR and two other groups

OpenAI will place evaluators from Apollo Research, METR, Transluce, and Goodfire inside the company in a pilot, with Apollo's starting this week, per TechCrunch. Six Apollo evaluators will investigate whether OpenAI's models act to undermine their own or their successors' alignmentAlignmentThe problem of making an AI system reliably pursue the goals its developers and users intend, and the research field devoted to it. Misalignment covers everything from a chatbot flattering users to a capable system deceiving or resisting its operators. and control mechanisms, the safeguards meant to keep models behaving as intended. Goodfire will help monitor training runs using its tools for examining how models work internally, and METR will look for dangerous behavior that OpenAI's own tests may have missed. OpenAI said the effort supports its commitment under the White House Accord on Super Intelligence to rely on independent assessments of safety controls. TechCrunch said limits on the evaluators' access and disclosure raise questions about their independence.

Read at TechCrunch ↗

Google finds attackers running AI exploit pipelines with no human in the loop

Google Threat Intelligence Group (GTIG) said in its quarterly AI threat report that, for the first time, it saw attackers use AI frameworks to fully automate finding and exploiting software flaws, per SiliconANGLE. One framework, exploit_agent, sat on an exposed server and drew on several frontier and locally run models. It compared popular web application code before and after each patch, then automatically ran the exploits it generated. Victim data suggests its operator targeted French media, entertainment and education websites. GTIG expects attacks run largely by AI agents with little human involvement within three months and sophisticated, fully autonomous ones within six, BleepingComputer reported. Suspected North Korean hackers also used Anthropic's Claude Code to deliver malware.

Read at SiliconANGLE ↗ • Read at BleepingComputer ↗ • Read at Google Cloud ↗

V.Industry & Market Watch

Southeast Asian data center lenders face loss of Chinese tenants under U.S. chip rule

Banks and private credit funds that lent billions for Southeast Asian data centers and AI chip purchases now risk losing Chinese customers, Bloomberg reported. Shares of data center and chip supply chain companies in the region fell after the U.S. named Singapore, Thailand, Malaysia and Indonesia as risks for diverting chips to China, per Reuters. Supparat Sivapetchranat Singhara Na Ayutthaya, vice chairman of the Thai Data Center Association, said roughly 300 megawatts of graphics processors aimed at Chinese users, costing about $12 billion to $18 billion, were due in Thailand over the next 12 to 18 months and could be redirected to U.S. customers. Data centers already serving Chinese companies have 90 days from the rule's effective date to obtain a license, according to CNA.

Read at Bloomberg ↗ • Read at Investing.com ↗ • Read at Channel NewsAsia ↗

OpenAI offers free classifier to flag AI generated child abuse images

At a press briefing in Washington, OpenAI announced a detection tool it says better identifies child sexual abuse imagery, including whether it was made with AI, and offered vetted organizations free access, per The Verge. The CyberTipline run by the National Center for Missing and Exploited Children received 400,000 reports tied to generative AI in 2025, up from 67,000 in 2024. OpenAI will also give developers detection guidance and train law enforcement on using AI in investigations. North Carolina Attorney General Jeff Jackson, co-chair of the Attorney General Alliance's AI Task Force, spoke at the event. Forty-eight states have laws addressing sexually explicit AI deepfakesDeepfakeSynthetic audio, images or video that realistically depict a real person saying or doing something they did not. Non-consensual intimate deepfakes and election deepfakes are the two areas where states and Congress have actually passed laws..

Read at The Verge ↗ • Read at OpenAI ↗

NATO taps OpenAI cyber models at up to 300 billion tokens a month

NATO's Communications and Information Agency (NCIA) has used OpenAI's cybersecurity models through Daybreak, the company's program for vetted defenders, consuming up to 300 billion tokens a month, UK Defence Journal reported, citing OpenAI. OpenAI has also given access to the EU cyber agency ENISA, which found four flaws in software that EU institutions rely on, and to national agencies in France, Germany and Poland. NCIA General Manager Dylan Browne said the agency will study usage and costs to "make informed procurement decisions." Ukraine's digital transformation minister, Oksana Ferchuk, said at NATO's Edge 26 conference that AI was making Russian cyberattacks larger and cheaper to mount.

Read at UK Defence Journal ↗ • Read at OpenAI ↗

VI.Global & Geopolitics

EU revives bid for international AI governance deal, eyes new U.S. Congress

The EU is reviving its bid for an international agreement on AI governance, betting that midterm losses for President Trump's party and public anger at tech companies improve the odds, per the Financial Times. Brussels has so far failed to persuade Trump to join an international AI governance pact, while warning that competition among AI companies is pushing out powerful models without enough safety testing. Asked about a new effort, European Commission President Ursula von der Leyen said it would be "very interesting to work with the Congress."

Read at Financial Times ↗ • Read at Stockwatch Cyprus ↗

UK actors' union secures consent and pay rights over AI digital replicas, pending member vote

Equity, the UK actors' union, reached a three year deal with the producers' body Pact giving performers the right to consent to, and be paid for, uses of their digital replicas, The Guardian reported. Producers may obtain permission only for a specific project rather than blanket consent, and may not use a replica during a strike where consent would be required. The deal goes to a vote of 8,000 Equity members, with union leaders recommending approval.

Read at The Guardian ↗