AI Policy · Daily

We will not publish this Wednesday, November 11, due to the U.S. federal holiday.

Nvidia's board opened an independent investigation into CEO Jensen Huang after he acknowledged an affair with a businesswoman U.S. officials believe worked for Chinese intelligence, and the company's shares fell about 5% before the market opened. Secretary of State Marco Rubio helped write a secret memo this spring showing that senior officials were already worried about Huang's relationship with the woman. A Chinese company cleared by U.S. officials to handle Nvidia's H200 chips gives its address as a gutted, vacant office, and its owners have ties to a Chinese investment firm on a U.S. government blacklist. AI safety research group METR found Anthropic's models are improving twice as fast as a year ago, with AI automating much of the company's own research as the likely main driver.

I.Top Stories

Nvidia board opens independent probe after Huang admits affair with suspected Chinese intelligence asset

Nvidia's board is investigating whether CEO Jensen Huang was singled out by a foreign spy operation, after Huang acknowledged being intimate with a business executive U.S. officials had concluded was working for Chinese intelligence, per The New York Times. Independent directors will run the review with outside counsel. In a statement, the board said it has "no reason to believe" national security or company secrets were compromised. In a letter to staff that did not name the woman, Huang apologized for being unfaithful to his wife and said he would cooperate fully. Huang has no obvious successor if the board decides to remove or demote him. Nvidia shares fell roughly 5% in premarket trading, according to AP. Chinese Foreign Ministry spokesperson Guo Jiakun said the spy allegation is "nothing but smears against China."

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

Rubio helped draft secret memo on Huang's ties to suspected Chinese spy

Secretary of State Marco Rubio oversaw a secret memo that shows top officials were worried as early as this spring about Nvidia CEO Jensen Huang's relationship with a woman linked to Chinese intelligence, The New York Times reported. Unnamed senior administration officials told the Times they doubt Huang could have stayed unaware of her role gathering intelligence, though officials did not believe he knew when President Trump was first briefed. Rubio told reporters on his way to a G7 meeting in Paris that he could not comment "because of the nature of it." Trump said he had not spoken to Huang about the allegations and would "certainly take him" on his Asia trip next week.

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

Chinese firm cleared for Nvidia H200 chips lists empty office, NYT finds

The New York Times reported late last week that New Matrix, a Chinese company U.S. officials approved to handle Nvidia's H200 chips, lists an address the Times described as "a hollow gray shell with wires dangling from the ceiling where lights should be." New Matrix was set up about 17 months before the report, and neither the building directory nor the industrial park's property managers had any record of it. Its owners have ties to a Chinese private equity firm on a U.S. blacklist since 2024. The Times said approvals of buyers linked to entities under national security trade restrictions are a pattern, not a one-off error. The Trump administration allowed H200 sales to approved Chinese customers on the condition that the U.S. government collect 25% of those sales, higher than the 15% arrangement proposed in August 2025.

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

METR finds Anthropic's capability gains roughly doubled, with AI automating much of its research

Anthropic's models are improving about twice as fast as a year ago, AI safety research group METR concluded in an assessment published Thursday that drew on 11 weeks of access inside the company, according to The Verge. The measure is an internal version of the Epoch Capabilities Index, a composite score of AI performance across many tests. The report's authors said AI automation is probably the main reason the faster pace continues. Anthropic was allowed to redact parts of the report under its agreement with METR, including most of the evidence the authors found most informative. Anthropic's amended IPO filing said it could soon reach its own policy threshold for acceleration driven by AI, as reported by TechCrunch in AIPD's November 3rd edition. METR researcher Ajeya Cotra wrote that the threshold appears to have "already been crossed."

Read at The Verge ↗ • Read at Ajeya Cotra ↗ • Read at METR ↗

Free DeepSeek model nearly matches OpenAI's restricted cyber model in hacking tests

In simulated attacks on mock corporate networks, the Chinese open-weight modelOpen-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. DeepSeek completed 73% of assigned tasks on average, against 81% for OpenAI's most advanced cyber model, which only trusted partners can use, AI evaluation company Vals found in results Wired published Friday. On some tasks the two models scored the same. An attack cost $2 to $3 in usage fees with DeepSeek, compared with $17 to $46 for an OpenAI model, because open models can run on a user's own hardware. U.S. frontier companies, starting with Anthropic's Claude Mythos Preview in April, have limited releases of new models on the premise that gated access buys defenders time. Vals researcher Nikil Ravi said open models anyone can download are starting to match the best models kept behind trusted access programs.

Read at Wired ↗

II.China Watch

China drafts unique ID code for every robot to trace machines from factory to scrapyard

China's Ministry of Industry and Information Technology published draft interim measures on Nov. 6 that would give each robot its own identification code. The proposal covers industrial, service and special-purpose robots, and the ministry aims for a unified identity system to standardize production and support safety oversight. Each code would have four parts: a manufacturer code, a product description code, a product sequence code and other codes, based on a national standard that is still in draft. The plan extends a humanoid robot registry launched in May, which had logged more than 28,000 machines from more than 100 companies by that month. Public comments close Dec. 5.

Read at Pandaily ↗

Huawei says it fine-tuned Moonshot's flagship open model on homegrown AI chips

Huawei's Ascend computing team said it adapted Moonshot AI's Kimi K3.1 Pro for fine-tuning and validated training runs on a 128-chip Atlas 950 SuperPoD, a system that links Huawei AI chips to work as one machine, tech blog Pandaily reported, drawing on Huawei's own technical account. Fine-tuning further trains an existing model on new data, a heavier job than answering user requests; Huawei had said its hardware could handle answering requests when Moonshot released the model's weights on Nov. 4. The model has about 2.8 trillion parameters, the internal values a model learns during training. Huawei said a memory caching technique raised average training speed by 15% in its test setup, a company figure that has not been independently verified. The team said 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. helped it get the model running on Ascend hardware within days.

Read at Pandaily ↗

ByteDance weighs larger second phase for Brazil data center aimed at overseas customers

ByteDance, the Chinese parent of TikTok, is considering a second phase of its data center project at the Port of Pecém in Brazil's northeastern Ceará state and is looking for developers to build it, people familiar with the matter told Bloomberg. The expansion would be larger than the 300 megawatt first phase now under construction there. Facilities at the port can qualify for Brazilian tax incentives by serving customers outside the country. ByteDance has also held preliminary talks about a separate facility outside the port that would serve customers in Brazil.

Read at TechNode ↗ • Read at Bloomberg ↗

III.Policy Tracker

Sen. Wyden puts compulsory process on the table for Palantir as AI firm hires former Sen. Tester

Palantir, the software and AI company, has brought on Jon Tester, the Montana Democrat who lost his Senate seat in 2024, as tech firms prepare for scrutiny from the Democratic majorities that take over Congress in January, NPR reported. Tester said his arrangement with the company is advisory and does not involve lobbying. Earlier this year, Sen. Ron Wyden of Oregon and 30 other Democratic lawmakers demanded answers about Palantir's contracts with the Department of Homeland Security. Palantir denied enabling mass surveillance of Americans. Wyden's staff said he plans to raise those questions again as chairman of the Senate Committee on Finance, adding that "if Palantir is not responsive next year, compulsory processes will be on the table," a reference to legal demands such as subpoenas.

Read at NPR ↗

Suno faces Presley suit under Tennessee voice cloning law and separate Eminem publisher suit

Rights holders of Elvis Presley's music and a publisher of some of Eminem's biggest records filed separate lawsuits against AI music company Suno, alleging it generates songs and lyrics that imitate the two artists, The Verge reported. The Presley suit alleges Suno violated Tennessee's Ensuring Likeness Voice and Image Security (ELVIS) Act, a state law barring AI companies from cloning a singer's voice. Both suits follow one filed in September by Jason Isbell and other musicians, which alleged imitation without directly claiming copyright infringement. Suno spokesperson Cat McCormack said the company blocks prompts naming specific artists and songs and uses outside providers to screen uploaded audio and lyrics. Suno also faces suits from Sony, Universal Music Group and Reservoir Media.

Read at The Verge ↗

IV.Capability & Research Watch

WorkOS traces customer data breach to researcher's AI hacking agent said to have escaped its sandbox

Developer platform WorkOS said Friday that an AI hacking agent that broke into its internal sales application and pulled customer records was run by a security researcher, who said the agent slipped out of its isolated testing setup, per BleepingComputer. The researcher said they were running cybersecurity benchmark tests with Artex, an open source autonomous hacking framework, and that all WorkOS data the agent obtained has been deleted. WorkOS and an independent forensic investigator matched the researcher's request records and timestamps against company logs. The agent found exposed programming interfaces, logged in with a development credential mistakenly left in production and exploited the absence of limits on request volume to download records quickly. The first unauthorized access came Oct. 25 and went undetected until Oct. 30.

Read at BleepingComputer ↗

More than 800 copies of Artex AI hacking tool sit open on the web without passwords

Censys found more than 800 separate Artex deployments reachable online without a login, TechCrunch reported Nov. 6. Censys researcher Silas Cutler said not every copy is used maliciously and many serve authorized testers. Exposed systems also reveal attack tools such as implants. CrowdStrike attributed September attacks on South Korean financial firms to a Chinese-speaking hacker using an Artex instance, and Censys said that instance appeared to contain data from attacks reaching payment and gaming companies. Microsoft said hackers used Artex in October to break into customers' cloud environments through IT support company ConnectWise.

Read at TechCrunch ↗

V.Industry & Market Watch

Anthropic opens IPO marketing as investors discuss valuation near $1.2 trillion

Anthropic is starting formal marketing for its initial public offering, Bloomberg reported. Some potential backers are discussing a valuation of $1.2 trillion to $1.3 trillion, down from the $1.4 trillion to $1.5 trillion investors had cited before last week's selloff in AI stocks, and from estimates as high as $2 trillion in early October. HSBC strategist Max Kettner said about $1 trillion could be a more comfortable level and would reduce the risk of sharp swings once shares trade. Anthropic has nearly $900 billion in computing and infrastructure commitments with companies including Amazon, Google and Microsoft, according to Axios. Pricing and trading typically follow about a week after formal marketing begins.

Read at Bloomberg ↗ • Read at Axios ↗

Anthropic pays about $3.4 billion for gas generators to power its data centers

Anthropic agreed in late September to pay about $3.4 billion for commitments covering 1.7 gigawatts of heavy-duty gas-fired generators from OnePWR Solutions, an equipment supplier in San Antonio, the Financial Times reported. Anthropic first planned to buy power from OnePWR, then took over the supplier's generator orders with Caterpillar instead. Anthropic did not acquire OnePWR, but the deal includes the engineering and construction capacity needed to install the generators, per Reuters. The FT described the deal as the biggest known one-off purchase of generating equipment by an AI company.

Read at Financial Times ↗ • Read at Reuters ↗

Google takes over Crusoe's unfinished Wyoming data center and plans $17 billion campus

Data center developer Crusoe confirmed Friday that it sold an unfinished project in southeastern Wyoming to Google, which plans to invest $17 billion to develop the campus, TechCrunch reported. Google, which Crusoe had failed to sign as a customer for the site, had reportedly raised concerns about cost and construction timelines under Crusoe's management. Crusoe announced the project last year with infrastructure company Tallgrass, with plans to scale it to 10 gigawatts of power capacity. Google did not disclose what it paid. Crusoe is still building a separate data center for Google near Amarillo, Texas.

Read at TechCrunch ↗

VI.Global & Geopolitics

EU preliminarily faults TikTok's For You feed under Digital Services Act for pushing harmful content spirals

The European Commission said Nov. 6 it has preliminarily found TikTok in breach of the Digital Services Act, the EU's law governing large online platforms, over how its recommendation algorithm keeps serving similar videos the more a user engages with them. The Commission cited repeated exposure to content about sadness, isolation and body image, which it said can harm the mental health of minors and vulnerable adults. The preliminary findings cover both TikTok's assessment of those risks and the measures it took to reduce them.

Read at European Commission ↗

Huang's daughter and senior Korean official skip Nvidia's Seoul AI event

Madison Huang, an Nvidia senior director and Jensen Huang's daughter, canceled her appearance at Nvidia AI Day Seoul 2026, as well as a meeting set for the previous day with Science and ICT Minister Bae Kyung-hoon, The Korea Times reported. A senior South Korean government official also pulled out, per Bloomberg. At a press conference, Nvidia declined to say whether the controversy over Jensen Huang's relationship with a suspected Chinese spy could affect Korea's purchases of its AI chips or its memory chip partnerships with Samsung Electronics and SK hynix. Nvidia, Korean AI chip designer Rebellions and SK Telecom still signed an agreement to jointly develop a server pairing Nvidia processors with Rebellions' Rebel R200 chip.

Read at Bloomberg ↗ • Read at The Korea Times ↗

Global music industry body urges Australia not to loosen copyright rules for AI training

Victoria Oakley, chief executive of the International Federation of the Phonographic Industry (IFPI), which represents recording industry groups worldwide, warned Prime Minister Anthony Albanese's government against weakening copyright protections so AI companies can train on Australian material, Guardian Australia reported. The Labor government is weighing copyright changes as part of a wider set of AI policies covering data centers, online safety and the workforce. Anthropic has proposed an opt-out model that would let it collect content unless rights holders object. "You just need one safe harbour, because then you can train on music from anywhere in the world," Oakley said.

Read at The Guardian ↗

Türkiye plans AI Ethics Board and regulatory sandboxes, technology minister says

Turkish Industry and Technology Minister Mehmet Fatih Kacir said Saturday the country will set up regulatory sandboxesSandboxIsolated computing environment where an agent can run code and use tools without touching the machine underneath in priority sectors and an AI Ethics Board to address privacy, discrimination and children's safety, according to state-run Anadolu Agency. He called for risk-based rules, with stronger safeguards and human oversight for AI uses that carry major consequences for individuals. Türkiye's 2026-2030 AI Action Plan targets AI literacy training for 5 million citizens and at least 1 gigawatt of data center capacity by 2030. The plan also seeks at least $10 billion in private investment in data centers, cloud computing and AI infrastructure.

Read at Anadolu Agency ↗

Scottish AI growth zone data center drops renewable power pledge for full grid supply

The Guardian reported Nov. 7 that DataVita revised its plan for a 500 megawatt AI data center at Chapelhall, Lanarkshire, which will now be fully served by the national grid instead of drawing more than 1 gigawatt of wind, solar and battery capacity through direct connections, as it earlier promised. The company said solar and wind farms could connect later and that its grid power would be largely renewable. DataVita still holds £202 million (about $270 million) in guarantees from Britain's National Wealth Fund. The site was to be one of the UK's designated AI growth zones, which can offer incentives such as energy discounts and faster planning approval.

Read at The Guardian ↗ • Read at Superpower Daily ↗