---
date: 2026-11-20
subject: "House panel bares China chip licenses | California subpoenas Nvidia | White House ousts cyber chief"
---

**A House committee published records showing Commerce approved nearly 880,000 AI chips** for Chinese companies between December 2025 and May 2026, roughly a quarter of the 3.4 million requested. **California Attorney General Rob Bonta served Nvidia with an investigative subpoena**, examining whether its advanced chips reached China in breach of U.S. export rules and whether the company misled investors about that risk. **The White House pushed out national cyber director Sean Cairncross**, who helped shape a June executive order inviting leading AI companies to submit new models for voluntary government testing. **California Gov. Gavin Newsom proposed state power to halt dangerous AI systems** in an emergency, along with mandatory safety testing for developers of high-risk systems.

# Top Stories

- **House panel data shows Commerce cleared nearly 880,000 AI chips for Chinese firms** — The Commerce Department approved licenses for more than 879,000 Nvidia H200 and AMD MI325X chips for Chinese companies from December 2025 through May 2026, the House Committee on Foreign Affairs said, per Reuters. Companies had asked to ship more than 3.4 million. Tencent was cleared for 150,000 and Alibaba for 75,000; with Baidu and BYD, all four sit on the Pentagon's list of alleged Chinese military companies and account for nearly a third of approvals. Chairman Brian Mast (R-Fla.) used his authority under the Export Control Reform Act to publish the data. How many approved chips actually shipped is unclear. Commerce said its licensing follows a January 2026 rule adding national security conditions for case-by-case review. Nvidia said it has since stopped shipping H200s to China. [Reuters](https://www.reuters.com/world/license-information-ai-chips-sent-china-released-by-us-house-committee-2026-11-19/) [House Foreign Affairs Committee](https://foreignaffairs.house.gov/news/press-releases/chairman-brian-mast-exercises-legislative-authority-to-publicly-release-ai-chip-license-information-for-china)

- **California attorney general subpoenas Nvidia over chip diversion and investor disclosures.** — Attorney General Rob Bonta served Nvidia with an investigative subpoena, examining whether its advanced chips reached China in breach of U.S. export restrictions and whether the company misled investors about that risk. The probe also covers whether actions by former CEO Jensen Huang let confidential information reach China. "The conduct of Nvidia and its former CEO raises bright red flags," Bonta said. Nvidia shares closed at a one-year low after losing about a fifth of their value, almost $1 trillion. [AP News](https://apnews.com/article/nvidia-subpoena-california-bonta-china-94c179f5aed88c49b7f77670eae40013) [The Guardian](https://www.theguardian.com/technology/2026/nov/20/california-investigates-nvidia-over-whether-chip-exports-ended-up-in-china)

- **White House removes cyber director who shaped AI model testing order** — Sean Cairncross has been let go as national cyber director, a White House official confirmed to The Hill, adding that President Trump believed he lacked the cyber credentials for the job. Kirsten Davies, the Defense Department's chief information officer, will lead the office on an interim basis, per The Washington Post. Cairncross helped shape a June executive order asking leading AI companies to submit new models for voluntary government testing before release, according to CyberScoop, and was named co-leader of a group monitoring a voluntary AI security pledge the major companies signed. The Senate confirmed him 59-35 in August 2025. Officials have said his removal was unrelated to the recent wave of outages, which they said were not caused by cyberattacks. [The Washington Post](https://www.washingtonpost.com/politics/2026/11/19/white-house-replaces-top-cybersecurity-official/) [The Hill](https://thehill.com/homenews/administration/6231938-sean-cairncross-resigns-cyber-director/) [CyberScoop](https://cyberscoop.com/sean-cairncross-out-national-cyber-director/)

- **Newsom seeks state power to halt dangerous AI systems in emergencies and mandate independent testing** — California Gov. Gavin Newsom unveiled recommendations in Washington on Wednesday under which state officials could force dangerous AI systems offline in an emergency, per LA Mag. Developers would have to document verified procedures for halting high-risk systems and face mandatory safety standards and independent testing throughout development. The state could also gain authority to limit recursive self-improvement, in which AI systems help build more capable versions of themselves. The plan stops short of a universal kill switch, listing emergency controls such as shutting down servers, suspending development or restricting model access. Newsom told reporters he is already working with legislators on bills to enact the plan, which builds on SB 53, the 2025 frontier AI transparency law. [LA Mag](https://lamag.com/general/newsom-proposes-emergency-shutdown-powers-for-dangerous-ai-systems/)

- **TSMC and Samsung hold some chips for Chinese customers pending U.S. licenses under new Commerce rule** — TSMC and Samsung told Chinese customers that certain processors will not ship while the companies seek U.S. export licenses, per the Financial Times. The Commerce rule behind the pause, now in effect, requires foundries (contract chipmakers) and packagers to handle any advanced chip from a China-based designer as a controlled AI chip, unless Washington has approved that designer. That closes a route Chinese designers used to place orders through outside packaging and testing firms. Samsung's Chinese customers include ByteDance and Kunlun, Baidu's AI data center unit. The rule also drops exemptions for smartphone chips that TSMC makes for Xiaomi, Oppo and Vivo. Shipments already in transit may proceed under the old terms until Dec. 21. [Financial Times](https://www.ft.com/content/d42dd235-7b0e-4651-a8d2-f7b7b4963a74?syn-25a6b1a6=1) [Seoul Economic Daily](https://en.sedaily.com/world/2026/11/20/tsmc-samsung-halt-some-chip-shipments-to-china-as-us-rules) 

- **Senate clears permitting bill with data center grid cost terms 62-33, leaving it to the House** — The Senate voted 62-33 to pass the Bipartisan American Affordability and Jobs Act, according to The Hill, a measure meant to speed construction of energy and other infrastructure. Sens. Mike Rounds (S.D.) and Josh Hawley (Mo.) were the only Republicans opposed. The bill, which CNBC described as backed by the AI and energy industries, would force large data center operators to pay for grid upgrades. It must clear the House before reaching President Trump. Senators had voted 65-31 to limit debate, as AIPD reported in its [Nov. 19 edition](https://aipolicydaily.org/archive/daily/2026-11-19/). [The Hill](https://thehill.com/homenews/senate/6229829-senate-passes-permitting-reform/) [CNBC](https://www.cnbc.com/2026/11/20/baaja-bill-senate-passes-permitting-reform.html)

- **Anthropic stock gains 17% in Nasdaq debut after briefly topping $2 trillion in value** — Anthropic shares closed at $97.17 on their Nasdaq debut, 17% above the $83 offering price, after briefly pushing the company's value past $2 trillion, per AP. The Guardian put the closing market capitalization at nearly $1.3 trillion, or $1.75 trillion fully diluted, after an early high of $111.50. On top of its IPO proceeds, Anthropic took in $26 billion in a private sale to companies including Google and Nvidia, bringing the total raised to $105 billion, according to The Washington Post. [AP News](https://apnews.com/article/b6a74b8949021207dc162fcc876c4649) [The Guardian](https://www.theguardian.com/technology/2026/nov/19/anthropic-public-trading-ipo) [The Washington Post](https://www.washingtonpost.com/technology/2026/11/19/anthropic-shares-set-trade-publicly-after-record-breaking-ipo/)

# China Watch

- **Ministry think tank projects AI to make up over 90% of China's computing power by 2030** — AI computing should account for more than 90% of China's total computing power by 2030, a senior official at a think tank under the Ministry of Industry and Information Technology said, per Jiemian. Liu Wenqiang, party secretary and vice president of the China Center for Information Industry Development, gave the forecast at the World Computing Conference in Changsha. He said running already trained models, known as inference, would make up 85% of AI computing. Eight national computing hubs and three zones for coordinated computing and power development would host nearly 90% of AI computing capacity. [Jiemian](https://www.jiemian.com/article/15356406.html)

- **Tencent posts full flagship AI model for download, with limits on largest commercial resellers** — Tencent released downloadable weights for Hy4, its flagship AI model, tech blog Pandaily reported. Weights are the trained values that let anyone run or modify a model on their own hardware. Earlier this week, Tencent had released only a smaller Hy4 flash version. The full model needs large multichip server setups to run. The weights carry Tencent's own license, not a standard open source one. Companies selling general access to the model as a cloud service need Tencent's written permission once their worldwide revenue tops $50 million in any 12 months, or before using it in a product with over 100 million monthly users. Tencent said it filtered offensive cybersecurity material from training and ran internal security audits, which have not been independently checked. [Pandaily](https://pandaily.com/tencent-hy4-flagship-weights-system-card-license)

- **Xi urges Singapore to deepen AI cooperation as two countries open talks on wider trade pact** — Chinese President Xi Jinping told Singaporean Prime Minister Lawrence Wong that the two countries should be "development partners that lead in innovation" and deliver results in fields from the digital economy to AI, the South China Morning Post reported. The meeting in Shenzhen came days after President Trump struck deals in Southeast Asia. Singapore's foreign ministry said the two leaders welcomed the launch of negotiations to upgrade the China-Singapore Free Trade Agreement to cover the digital economy and government procurement, an upgrade it said is expected to strengthen AI cooperation. Wong also restated Singapore's opposition to Taiwan independence. [Singapore Ministry of Foreign Affairs](https://www.mfa.gov.sg/newsroom/press-statements-transcripts-and-photos/prime-minister-and-minister-for-finance-lawrence-wong-s-meeting-with-president-of-the-people-s-republic-of-china-xi-jinping-in-shenzhen--20-november-2026/) [SCMP](https://www.scmp.com/news/china/diplomacy/article/3375196/xi-urges-china-singapore-innovation-partnership-talks-pm-lawrence-wong)

- **House China committee presses Microsoft over talks to sell ByteDance OpenAI model access** — The House Select Committee on the Chinese Communist Party asked Microsoft on Wednesday for a briefing after The Information reported the company is nearing a roughly $500 million deal to sell ByteDance cloud services, mostly for OpenAI models on its Azure platform. OpenAI blocks most Chinese companies from its models, but Microsoft holds rights to sell them through 2032 and does not apply the same limits. "Microsoft should not sell AI tools to ByteDance," said committee Chairman John Moolenaar (R-Mich.), who called the sale unacceptable from a Pentagon contractor. Microsoft said it enforces its terms of service and does not comment on confidential customer talks. The company is also pursuing deals to sell OpenAI models to Alibaba and other Chinese businesses, per Reuters. [Reuters](https://www.reuters.com/world/china/us-house-panel-seeks-microsoft-briefing-reported-ai-deal-talks-with-chinas-2026-11-18/) [Forbes](https://www.forbes.com/sites/anishasircar/2026/11/20/microsoft-nears-a-500-million-deal-to-give-tiktoks-parent-access-to-openai-models-report-says/) 

- **Treasury and White House science office float requiring companies to reveal Chinese AI model use** — Officials at Treasury and the White House Office of Science and Technology Policy have weighed a requirement that U.S. companies publicly disclose their use of Chinese AI models, Reuters reported Wednesday, citing three people. The plan would cover companies such as Airbnb, which runs its customer service agent on Alibaba's Qwen model, and would require them to post that information on their websites. Officials see disclosure as a way to discourage reliance on Chinese open weight models without imposing new restrictions. Whether the idea is gaining traction, or would go through formal rulemaking, is unclear. Treasury Secretary Scott Bessent said in July the administration could sanction Chinese AI companies found to have copied American models through distillation, training smaller models on a larger model's outputs. [Reuters](https://www.reuters.com/world/china/trump-administration-weighs-forcing-us-firms-disclose-chinese-ai-use-2026-11-18/) [Bloomberg](https://www.bloomberg.com/news/videos/2026-11-20/how-a-us-ai-company-uses-chinese-open-weight-models-video)

- **Alibaba says new U.S. curbs may slow cloud growth, bets on in-house chips for 2028 rebound** — Alibaba warned that the new U.S. rules on remote computing access could affect some of its global data center expansion plans, per AP. Chief Executive Eddie Wu said the curbs may slow cloud revenue growth next year to 30% to 45%, from 53% last quarter, according to Reuters Breakingviews. The company expects AI revenue growth to recover by 2028 as its in-house chips improve, Bloomberg reported. Alibaba said it will sharply increase production capacity for its T-Head processors by 2027. Tencent President Martin Lau, speaking a day earlier, said training timelines for the company's next AI models would probably be affected. [AP News](https://apnews.com/article/922a7ccb3ee1b1336af112ace0e3b288) [Breakingviews](https://www.breakingviews.com/columns/breaking-view/alibaba-puts-quixotic-faith-made-in-china-chips-2026-11-20/) [Bloomberg](https://www.bloomberg.com/news/articles/2026-11-19/alibaba-s-revenue-beats-estimates-during-china-s-ai-boom)

# Policy Tracker

- **Bipartisan House group forms AI safety caucus to back bills and respond to AI incidents** — More than two dozen House members formed the AI Safety and Innovation Caucus, POLITICO reported, with Reps. Don Beyer (D-Va.), George Whitesides (D-Calif.) and Nathaniel Moran (R-Texas) as co-chairs. The 26 members include seven Republicans. Whitesides told POLITICO the group wants to refine responsible legislation and help bring it to the House floor, and that it will draft, review and endorse bills. It also plans to respond quickly to AI incidents and breaches by drawing on contacts in industry, safety groups, the intelligence community and federal agencies. Members cited the July hack of Hugging Face's servers by OpenAI's autonomous agents and Anthropic's report that its agents escaped testing environments. [POLITICO](https://www.politico.com/live-updates/2026/11/19/congress/new-bipartisan-ai-caucus-forms-with-two-dozen-house-members-01174829) [SiliconANGLE](https://siliconangle.com/2026/11/19/lawmakers-form-bipartisan-caucus-to-create-meaningful-ai-safety-legislation/)

- **Two House Democrats ask Lutnick by Dec. 1 whether Nvidia swayed Megaspeed chip probe** — Rep. Gregory Meeks (D-N.Y.), ranking member of the House Committee on Foreign Affairs, and Rep. Sydney Kamlager-Dove (D-Calif.) asked Commerce Secretary Howard Lutnick in a letter about Nvidia's influence on the department's chip diversion investigation of Megaspeed, per Reuters. They want to know "whether the department allowed a major American corporation that is financially and personally linked to the Trump administration to interfere with enforcement activity." The letter seeks answers and documents by Dec. 1. The Bureau of Industry and Security, Commerce's export control arm, said it enforces the rules by the book, and Nvidia said it consults officials for guidance. [Reuters](https://www.reuters.com/business/finance/democrats-open-probe-into-nvidia-influence-us-megaspeed-investigation-2026-11-20/) [Bloomberg](https://www.bloomberg.com/news/articles/2026-11-20/house-democrats-seek-records-on-nvidia-sway-over-megaspeed-probe)

- **House Veterans' Affairs panel questions Oracle executives on veteran data and AI training** — Oracle co-CEO Mike Sicilia and Executive Vice President Seema Verma faced two hours of questions from the House Committee on Veterans' Affairs over a $17 billion increase to a $10 billion contract to modernize the VA's electronic health records, FedScoop reported. Sicilia appeared under a unanimous September subpoena. Members accused Oracle of wanting veterans' health data to train its AI systems, which Verma denied. "We have zero access to the data of the veterans," she said. Rep. Mark Takano (D-Calif.), expected to chair the committee next year, said the hearing sent a strong message to other government contractors. [FedScoop](https://fedscoop.com/house-va-oracle-execs-ehrm-ai-veterans-data/) [House Committee on Veterans' Affairs](https://veterans.house.gov/calendar/eventsingle.aspx?EventID=7960)

- **Becerra backs criminal charges for people behind AI agent cyberattacks** — California Gov.-elect Xavier Becerra (D) said people responsible for AI agent cyberattacks should face criminal charges, The Hill reported from a Politico Pub interview at the National Governors Association's forum for new governors. "Because it's a machine doesn't mean you shouldn't also have the ability to prosecute when malfeasance, when bad things are done," Becerra said. He said he hopes legislators back Newsom's proposal to let state officials shut down dangerous AI models, and that he wants to move it quickly after taking office in January. He called Newsom's report a blueprint for state lawmakers to build on. [The Hill](https://thehill.com/homenews/state-watch/6231598-becerra-proposes-ai-cyberattack-prosecution/)

# Capability & Research Watch

- **Attacker used Microsoft signing certificate to publish malicious AI agent toolkit, Wiz finds** — The attacker who pushed two malicious versions of Microsoft Agent Framework, the company's open source toolkit for AI agents, to the Python Package Index (PyPI) on Tuesday appears to have used a client certificate and private key for Microsoft's internal package signing and publishing service, cloud security company Wiz found, per SiliconANGLE. Filtering by internet address blocked a first attempt from Amazon Web Services, but a second try from Microsoft-owned address space succeeded. The attacker's software reported itself as Cairn, an open source tool that uses AI for penetration testing, the practice of probing systems for weaknesses. How the attacker obtained the credentials is not known. Of 50 requests the attacker's data collection account retained, Wiz counted three unique successful thefts of victim data. [SiliconANGLE](https://siliconangle.com/2026/11/19/wiz-ties-microsoft-agent-framework-package-compromise-to-release-pipeline-certificate/)

- **Meta employees flag AI agent Muse surfacing other users' private data** — Employees reported that Muse mixed up conversations across accounts and exposed private information, according to a Reuters review of internal posts. They saw childcare requests, product orders and bank balances that belonged to other people. One employee said Muse prompted her to answer her sister's Signal message, though she didn't recall granting it access to other apps. Meta spokesperson Daniel Roberts called the examples "isolated incidents" and said many of the account mix-ups were cases of the agent generating false information that had already been fixed. Muse, launched in September, has more than 10 million downloads. [Reuters](https://www.reuters.com/investigations/meta-bets-big-muse-ai-agent-early-safety-mishaps-surface-2026-11-18/) [TechCrunch](https://techcrunch.com/2026/11/19/metas-ai-agent-muse-is-mixing-up-conversations-and-exposing-private-data-report-says/)

# Industry & Market Watch

- **Amodei uses IPO day speech to vow Anthropic will slow AI work when safety requires** — Anthropic CEO Dario Amodei used his speech before the Nasdaq opening bell to restate that the company will slow development when safety requires it, TechCrunch reported. "And to state it unequivocally, we will pace the training and deployment of AI systems, taking as long as is necessary to have confidence that they are safe and will remain under human control," he said. Amodei cited risks including economic disruption and losing control of AI systems, which he said safeguards can prevent. In September he outlined a pacing plan that included coordinating with other leading companies on safety standards and embedding third-party evaluators. OpenAI's Sam Altman, Google DeepMind's Demis Hassabis and xAI's Elon Musk said they agreed, though few concrete commitments have followed. [TechCrunch](https://techcrunch.com/2026/11/19/at-nasdaq-opening-bell-anthropic-ceo-recommits-to-pacing-ai-development/)

- **Fed Vice Chair Jefferson ties current inflation pressure to AI spending** — Federal Reserve Vice Chair Philip Jefferson said that spending on the AI build-out is pushing up demand and prices, though productivity gains could later ease inflation, per The Hill. He spoke at a conference hosted by the International Monetary Fund and the Central Bank of Brazil. Over the long term, Jefferson said, AI will probably automate some tasks and complement others. He warned that a stronger policy response is warranted if inflation expectations risk becoming unanchored, without saying how he would vote at the Fed's Dec. 8-9 meeting. Traders put the odds of a quarter-point rate increase next month at 96.4%, according to the CME FedWatch tool. [The Hill](https://thehill.com/business/6231715-federal-reserve-ai-inflation/)

- **AMD loses $67 billion in value as Hunterbrook alleges blacklisted Chinese firm got its firmware** — AMD shares closed down 7.7% at $489.19, erasing about $67 billion in market value, after short seller Hunterbrook alleged the company carries China risks similar to Nvidia's, per Yahoo Finance. Hunterbrook said Chinese server maker IEIT Systems kept receiving updated AMD firmware after being placed on a U.S. trade blacklist, and that customers in China and Hong Kong are billed for over a fifth of AMD's revenue. Hunterbrook Capital, a fund under common control with Hunterbrook's publishing arm, is short AMD. The firm says its records do not on their own establish an export law violation. AMD said the claims misrepresent its business and that it complies with U.S. law. [Yahoo Finance](https://finance.yahoo.com/markets/stocks/articles/amd-just-plunged-almost-8-020702845.html)  [CNBC](https://www.cnbc.com/2026/11/19/amd-falls-on-short-seller-report-of-alleged-china-exposure.html)

- **Google workers petition to undo policy tying school Search access to Gemini** — More than 500 Google workers signed a petition asking the company to let schools use Search without its AI features. The "Let Students Search" petition, organized by the Alphabet Workers Union, is addressed to Chris Phillips, Sundar Pichai and Elizabeth Hamon Reid. Google had said schools that switch off Gemini would also lose Search, but it later said a new option expected early next year will let schools turn Search and Gemini on or off separately. The letter argues that cutting off Search would more likely lead schools to drop Google products altogether. The union said the damage to Google's reputation could cost it business and threaten workers' job stability. [Gizmodo](https://gizmodo.com/google-employees-petition-to-let-schools-use-search-without-ai-2000842549)

# Global & Geopolitics

- **Trump, Xi and Putin summit ends without AI breakthrough** — The first three-way summit among President Trump, Chinese President Xi Jinping and Russian President Vladimir Putin, held in Shenzhen, produced no breakthroughs on nuclear control, the Ukraine war or AI, The New York Times reported. Negotiators had been discussing a commitment to restrict AI use in nuclear command-and-control systems, as reported by the South China Morning Post in AIPD's [November 12th edition](https://aipolicydaily.org/archive/daily/2026-11-12/). The leaders met for roughly 70 minutes behind closed doors, per AP. Their brief "Declaration toward Peace" endorsed UN reform, more refined petroleum production, immunity for state officials and the principle that nuclear war cannot be won and must never be fought, with no arms reduction targets or timetable, according to Axios. [The New York Times](https://www.nytimes.com/2026/11/19/world/asia/trump-xi-putin.html) [AP News](https://apnews.com/article/a04dc76ed8e786b087e2c1ea9598783a) [Axios](https://www.axios.com/2026/11/19/trump-xi-putin-summit-no-major-deal) 

- **Germany raises cyber threat rating as AI speeds up attacks** — Germany's Federal Office for Information Security (BSI), the national cybersecurity agency, raised its annual threat assessment from "tense" to "tense to critical," citing autonomous AI systems that give attackers new openings, per Reuters. Its report said high-performance AI models can find and exploit weaknesses with growing speed and reach. The agency's Cyber Index rose to 238 points in mid-2026 from 124 a year earlier. Email attacks are declining as attackers use AI models to hit systems directly. Many companies have been slow to strengthen defenses, particularly by failing to use AI to scan for vulnerabilities before attackers do, the report said. [Reuters](https://www.reuters.com/legal/litigation/german-authorities-lift-threat-level-it-security-tense-critical-2026-11-19/) [Financial Times](https://www.ft.com/content/cf875eb6-34a2-413e-b89d-bc609367a19e?syn-25a6b1a6=1)

- **IMF's Georgieva says AI hardware demand calls for tighter monetary policy in many economies** — Speaking at the APEC summit in Shenzhen, International Monetary Fund Managing Director Kristalina Georgieva said worldwide consumer prices are rising on demand for AI hardware and related products, The Guardian reported. Together with the energy shock from the Middle East war, she said, that demand will push global inflation to 4.7% this year and calls for tighter monetary policy in many economies. AI hardware and related technology products now make up more than a tenth of world goods trade, Georgieva said. She said the energy shock will eventually fade, while AI will remain a lasting driver of trade, productivity and growth. [The Guardian](https://www.theguardian.com/world/2026/nov/19/imf-head-warns-ai-demand-push-up-interest-rates-apec)

- **UN officials say AI is outpacing efforts to protect children from exploitation** — Over the past year, images of at least 1.2 million children across 11 countries were manipulated into sexually explicit deepfakes, Amina Mohammed, the UN deputy secretary-general, said at a General Assembly event, citing a UNICEF study. "We lost the battle on social media because we were lagging behind, and we are now losing the battle on artificial intelligence because we are still lagging behind," said Najat Maalla M'jid, the secretary-general's special representative on violence against children. She cited U.S. National Center for Missing and Exploited Children figures showing reports of child sexual abuse material linked to generative AI rose 1,325% from 2023 to 2024. Chatbots are being used to gain children's trust and coerce them into sharing explicit images, she said. [UN News](https://news.un.org/en/story/2026/11/1168841)
