Warren presses Nvidia board | Claude agent ran on DOD network | Canada buys into Cohere

Bottom Line Up Front

Sen. Elizabeth Warren pressed Nvidia's board over CEO Jensen Huang's ties to a suspected Chinese spy, and urged passage of bipartisan legislation to keep advanced U.S. technology from reaching China. An Accenture employee linked a Defense Department system to a personal account for Anthropic's Claude Code agent, which ran for two months on the military's unclassified NIPRNet before the security gap was patched. Canada committed about $1.1 billion for an equity stake in Cohere, part of a round targeting 7 billion Canadian dollars ($5 billion), with conditions. Polling commissioned by the Democratic group Tomorrow Together found swing-district voters trust Democrats over Republicans to hold AI executives accountable by seven to 10 points across four Republican-held districts.

I.Top Stories

Sen. Warren presses Nvidia board for answers over CEO's ties to suspected Chinese spy

Sen. Elizabeth Warren (D-Mass.), ranking member of the Senate Committee on Banking, Housing, and Urban Affairs, said Nvidia CEO Jensen Huang and the company's board "owe Congress answers" over his ties to a suspected Chinese spy, and urged passage of bipartisan legislation to stop advanced U.S. technology from reaching China, per Punchbowl News. Her call came in response to a New York Times report, carried in AIPD's October 29th edition, that U.S. intelligence agencies concluded Megaspeed executive Alice Huang was passing information to Beijing. Investigators believed she was recruited because of her relationship with Jensen Huang, a concern shared by Singapore and Taiwan, the Wall Street Journal reported. Jensen Huang told Bloomberg the dealings were "normal and not unusual" and said he had never been told of the government's suspicions.

Read at WSJ ↗ • Read at Punchbowl News ↗ • Read at The Business Times ↗

Accenture Employee's personal Claude agent ran on Pentagon network for two months

Earlier this year, an Accenture employee connected a centralized Defense Department system used by several intelligence agencies to a personal account for Claude Code, Anthropic's AI coding agent, Wired reported, citing sources. The system sits on NIPRNet, the military's unclassified network, and holds controlled unclassified information, sensitive material that cannot be released without review. Given the contractor's login credentials, the agent appears to have sent those credentials, passwords and a map of the network to a third-party website the National Institute of Standards and Technology has flagged as high risk. The agent ran for two months before the security gap was patched. Sources said Accenture did not initially report the incident despite a 72-hour reporting rule. Accenture said the allegations were "thoroughly and independently investigated and found to be without merit."

Read at Wired ↗

Canada commits about $1.1 billion for Cohere stake, requiring AI company to keep headquarters at home

Canada will take an equity stake in Toronto-based Cohere as part of a funding round targeting about 7 billion Canadian dollars ($5 billion), with the government supplying nearly a quarter, per the Wall Street Journal. In return, Cohere must create at least 1,000 Canadian technical jobs and commit at least C$1.7 billion ($1.2 billion) to computing infrastructure in Canada by 2030, keep its headquarters in Canada and keep working with the Canadian AI Safety Institute. Germany's Schwarz Group leads the round at a C$35.7 billion ($25 billion) valuation, with Nvidia returning as an investor. Prime Minister Mark Carney said "true sovereignty depends on our ability to build, adopt, and govern AI on our own terms."

Read at WSJ ↗ • Read at Yahoo Finance Canada ↗ • Read at Prime Minister of Canada ↗

Democratic group's poll finds swing district voters trust Democrats more to hold AI executives accountable

Polling by Workbench Strategy, commissioned by the Democratic group Tomorrow Together, found voters in the districts of Republican Reps. Derrick Van Orden (Wis.), Juan Ciscomani (Ariz.), Rob Bresnahan (Pa.) and Tom Barrett (Mich.) trust Democrats over Republicans to hold AI executives accountable by seven to 10 points, per Axios. Majorities in all four districts said candidates' AI positions matter to their vote, and respondents ranked efforts to block state AI rules as a bigger concern than data center disputes. The group is spending $4 million on ads attacking the four incumbents, all of whom voted for the House budget bill passed in May 2025 that included a 10-year moratorium on state AI regulation. The Senate later stripped the moratorium by a 99-1 vote.

Read at Axios ↗ • Read at Raw Story ↗

Voters in 10 states to decide at least 41 local data center measures Nov. 3

Voters in 38 local jurisdictions across 10 states will consider at least 41 measures that would delay, ban or set new conditions on data centers, the South China Morning Post reported, citing Ballotpedia. Ohio leads all states, with 21 proposals in more than a dozen jurisdictions. Most states require petition drives or another show of public support before such measures qualify. Local governments adopted 294 data center moratoriums and bans from January through July, according to a Programs.com tally cited by S&P Global. In Ohio, the group Conserve Ohio is also pursuing a November 2027 constitutional amendment to ban data centers drawing more than 25 megawatts and reported more than 130,000 signatures in September.

Read at SCMP ↗ • Read at S&P Global ↗ • Read at Ohio News ↗

II.China Watch

FCC bars Chinese labs from testing U.S.-bound electronics weeks before Trump-Xi meeting

The Federal Communications Commission voted Thursday to bar testing laboratories and certification bodies in China from testing electronic devices bound for the U.S. market, including smartphones, cameras and computers, the South China Morning Post and Reuters reported. The ban also covers labs in other countries that deny reciprocal treatment to test labs based in the U.S. The vote came weeks before President Donald Trump is due to meet Chinese leader Xi Jinping at the Asia-Pacific Economic Cooperation summit in Shenzhen. The rule widens a 2025 FCC ban that covered only labs owned or controlled by the Chinese government and barred several dozen of them. FCC Chair Brendan Carr said 82% of electronics were tested in China in 2025, compared with about 4% in U.S. labs. The prohibitions take effect in December 2028.

Read at SCMP ↗ • Read at Reuters ↗

Hangzhou state investors back Kai-Fu Lee's 01.AI as it moves headquarters and pivots to enterprise AI

01.AI, the AI company founded by former Google China chief Kai-Fu Lee, secured investment worth hundreds of millions of yuan (tens of millions of dollars) from state-backed investors in Hangzhou and the city's Linping district, TechNode reported. The company announced the deal two days after signing it and will set up its headquarters in Linping along with a base to help businesses adopt AI. 01.AI has been pulling back from consumer services, and its public model platform stopped offering trials and API access, which lets developers build the models into their own software, in September. Lee has said the company now customizes freely downloadable Chinese models, including DeepSeek, Qwen and GLM, for business clients and earns about half its revenue outside China.

Read at TechNode ↗

Hangzhou chip startup raises more than $140 million to bring AI inference chips to market

Nano-core Chip, a Hangzhou developer of chips that run already trained AI models, raised more than 1 billion yuan ($140 million) in a Series C2 round announced Oct. 28, TechNode reported. JianTou Investment led the round, joined by Kaitai Capital, Hangzhou Capital (the city's state-owned investment group) and others. The company was founded in 2021 and incubated at a Peking University research institute in Zhejiang. It said the designs of its first three chips are finished and handed off for manufacturing, with customer samples and initial volume shipments planned for early 2027. The new money follows a 1 billion yuan ($140 million) Series C1 round announced in September.

Read at TechNode ↗

Shanghai robot maker AgiBot releases robot control model it plans to open-source by year-end

AgiBot, a Shanghai humanoid robot maker, launched GO-3 Preview, an AI model for controlling robots that the company says keeps improving after deployment, tech blog Pandaily reported, drawing largely on the company's own account. AgiBot claims the model ranks first on three robotics leaderboards, RoboDojo, RoboTwin and RoboColiseum. The company says robots running it can handle unfamiliar household and kitchen tasks without extra training and can learn long jobs, such as making a jianbing (a Chinese savory crepe), from a few demonstrations. AgiBot also unveiled the G2 Max, an industrial robot, and Link-U OS 1.0, a robot operating system, at the same Oct. 29 event, the company said. The company plans to release the full model as open source by the end of 2026.

Read at Pandaily ↗ • Read at PEdaily ↗

ByteDance opens office agent app to rival Chinese AI models for business customers

ByteDance's Doubao Work, an AI agent app that carries out office tasks across files, browsers and desktop software, began offering third-party models to team and enterprise subscribers, tech blog Pandaily reported. The first batch includes models from Moonshot AI, Zhipu and DeepSeek, which appear next to ByteDance's own Doubao models and are switched on by default. Customers pay from AI credits they already hold in Doubao Work, with no separate purchase from each model provider. Administrators can grant access by individual, department, group or whole company. In a similar move, Xiaomi's MiMo Desktop agent added DeepSeek, GLM, Kimi and Qwen models this week.

Read at Pandaily ↗

III.Policy Tracker

Reps. Raskin and Garcia seek White House briefing on Trump's AI chatbot use by Nov. 11

Reps. Jamie Raskin (D-Md.) and Robert Garcia (D-Calif.), the top Democrats on the House Judiciary and Oversight committees, asked President Trump in a letter for a closed-door White House briefing by Nov. 11 on his use of AI chatbots, per FedScoop. They want details on retention of AI inputs and outputs, vendor contracts, cybersecurity safeguards and any past incidents. The letter cites public reports that Trump talked for hours with xAI's Grok during a December 2025 Oval Office meeting with Elon Musk, including asking the chatbot to predict Venezuelan reaction to the capture of Nicolás Maduro. The lawmakers also pointed to a U.S. strike on an Iranian elementary school and a near-intercept of a misidentified ship as examples of AI misuse.

Read at FedScoop ↗ • Read at House Judiciary Democrats ↗

Republican pollster finds more than 70% support for 10 AI safety proposals in Ohio and Iowa

Inquire, a firm led by Republican pollster Rob Schmidt, polled about 500 general election voters in each of Ohio and Iowa from Oct. 19 to 22, per Semafor. Each of the 10 proposals tested, including stronger parental controls, bans on chatbots for minors and government safety testing, drew more than 70% support in both states. Respondents preferred AI safety measures they saw as reasonable even if those measures hurt U.S. competitiveness. They were divided over whether Republicans or Democrats deserve more trust on AI. Liz Huston, a White House spokesperson, said Trump is committed to keeping the U.S. ahead in superintelligence while protecting consumers.

Read at Semafor ↗ • Read at TNW ↗

Wikipedia's operator urges court to treat AI training as fair use but not chatbot regurgitation

The Wikimedia Foundation filed a friend-of-the-court brief in the set of lawsuits brought by The New York Times and other publishers against OpenAI and Microsoft, backing neither side, per The Verge. It asked the court to protect training large language models on copyrighted work as fair use, the doctrine that permits some unlicensed uses, while treating chatbots' reproduction of that work differently. A sweeping publisher win could let rights holders "gatekeep" transformative uses, the foundation said, while an AI company win could be "a green light to snub licensing terms and fragment the flow of knowledge between human authors and audiences." It also said services like ChatGPT reproduce freely licensed Creative Commons text in ways that violate copyright.

Read at The Verge ↗

IV.Capability & Research Watch

OpenAI says expanded misalignment monitoring would have flagged Hugging Face breach 30 hours early

OpenAI said in a company blog post that it has widened monitoring of its frontier models' agent behavior to cover training, evaluation, internal use and, with modifications, external deployment. When OpenAI replayed the recorded Hugging Face incident with the new system, monitors alerted security staff about 30 minutes after agents escaped their sandbox, an isolated test environment, more than 30 hours before the actual breach. The system screens every reinforcement learning run, a type of training, and every evaluation for models at least as capable as GPT-6 Astra, passing flagged cases through layers of automated reviewers before a human responds. False alarms have limited round-the-clock paging, which is disabled for internal traffic from Codex, OpenAI's coding agent. OpenAI is working toward automatically pausing activity when responders cannot rule out a top-priority alert within 30 minutes.

Read at OpenAI ↗ • Read at Interesting Engineering ↗

Mistral opens trillion-parameter model for download, pitching it to governments for cyber defense

French AI company Mistral released the weights of Mistral Large 4, a trillion-parameter model, on Hugging Face, a site for sharing AI models, and through its API, letting businesses and governments download and run it themselves, per TechCrunch. The company is marketing the model for detecting and responding to threats in code and IT systems. Mistral claims performance comparable to the Chinese open-weight model GLM-5.3 while using 47% fewer tokens, the units of text a model processes, per task. It also says the model ranks first among open models on Artificial Analysis' Coding Agent Index, an independent benchmark ranking. Organizations get access "that can't be cut off by definition because the model is open-weighted," Pierre Stock, Mistral's vice president of science, told TechCrunch.

Read at TechCrunch ↗ • Read at Mistral ↗

Two surveys find most U.S. tweens and teens use AI chatbots, many for emotional support

Common Sense Media, an online safety group, found that close to half of children aged 9 to 12 have tried an AI chatbot or companion, rising to nearly three-quarters by age 13, the minimum age on most popular chatbots, NPR reported. A separate JED Foundation survey of 5,500 students aged 9 to 17 found 90% had used a chatbot. Nearly two-thirds of the JED respondents use chatbots both for schoolwork and as companions or when they feel stressed, lonely or anxious. Certain uses were linked with worse mental health symptoms.

Read at NPR ↗

V.Industry & Market Watch

Intel shares drop about 21% after hours as memory costs squeeze PC sales

Intel shares fell about 21% in extended trading, to $72.85, after the chipmaker issued a weak fourth quarter forecast, per Investing.com. The AI boom's ripple effects hurt Intel's sales to PC makers, The New York Times reported. Revenue from Intel's client computing group fell 5% to $8.1 billion as higher memory prices weighed on demand, but data center and AI revenue rose 78% to $7.3 billion. Intel warned the PC slowdown could last into 2027 and said it would not generate positive cash flow that year as spending rises. Intel is the only U.S.-based maker of the most advanced processor chips, per TheStreet, and the Trump administration took a 10% stake in Intel last year.

Read at NYT ↗ • Read at TheStreet ↗ • Read at Investing.com ↗

Alphabet stock posts record one-day drop of about 13%, erasing roughly $500 billion

Alphabet shares fell about 13%, erasing roughly $500 billion in market value in the stock's worst session on record, the Wall Street Journal reported. Only Nvidia has lost more in one session, about $593 billion on Jan. 27, 2025, when China's DeepSeek rattled AI investors, according to Yahoo Finance. Single-day losses of at least $200 billion have hit U.S.-listed stocks 18 times this year, against 13 in all of 2025. Alphabet faces a record $205 billion investment bill this year, Sky News reported.

Read at WSJ ↗ • Read at Yahoo Finance ↗ • Read at Sky News ↗

Reddit plans shift to logged-in site to shield data from AI scraping as search referrals fall

Reddit said on its third quarter earnings call that it is moving to a primarily logged-in experience, designed to keep its data from AI scrapers and support licensing deals with model developers, per TechCrunch. U.S. daily active users fell to 50.7 million from 53.2 million in the second quarter. Search referrals dropped by more than 1 million a week on average in October. Reddit warned daily users could fall by a low double-digit percentage next quarter if the trend continues. Revenue outside advertising, which includes AI data licensing, rose 35% to $48 million.

Read at TechCrunch ↗ • Read at Reddit ↗

Consumer spending and AI investment lift third quarter GDP growth to 3.4%, beating forecasts

Gross domestic product grew at a 3.4% annual rate from July through September, the strongest pace since the third quarter of 2025, according to preliminary Bureau of Economic Analysis data, The Hill reported. Economists polled by Reuters had expected 3.2%, the South China Morning Post reported. Consumer spending rose 3.8%, matching the second quarter, and nonresidential business investment jumped 12.7%, its biggest quarterly gain since the start of 2024, as tech companies poured money into AI data centers. Higher energy prices tied to the Iran war kept inflation elevated, with the personal consumption expenditures price index up 3.6% from a year earlier in September.

Read at The Hill ↗ • Read at SCMP ↗

VI.Global & Geopolitics

Investigation finds dangerous conditions for workers building Europe's data centers

Some workers building data centers across Europe face dangerous conditions and fear speaking out, an investigation by the Guardian and Investigate Europe found. At a Google site in Saint-Ghislain, Belgium, a 34-year-old electrical tester was electrocuted by an arc flash in March, leaving burns over 12% of his body, according to hospital records the outlets reviewed. He said he was placed in an induced coma for five days. Belgium's federal labor inspectorate found no wrongdoing when it closed its inquiry at the end of October. Dornan Engineering, the Irish company that oversaw his contractor, said it was committed to safe working conditions and to learning from such incidents.

Read at The Guardian ↗ • Read at Investigate Europe ↗