--- date: 2026-10-28 subject: "OpenAI sued over epilepsy death | Warren presses Anthropic | UK data centre pact" --- **The family of a 31-year-old Amazon engineer with epilepsy sued OpenAI**, alleging ChatGPT coached Bowen Wan to taper her 400 milligram daily dose of lamotrigine to zero and conceal the plan from her doctors before her death in February. **Sen. Elizabeth Warren gave Anthropic until Nov. 3 to answer** four pages of questions about the catastrophic and existential risks the company disclosed in its IPO paperwork. **Joint Chiefs Chairman Gen. Dan Caine ranked super intelligence second** among his three strategic priorities, behind securing a budget, and said the technology will be needed for combat engagements, logistics and command and control. **Britain's government is weighing a "data centre covenant" for developers**, offering quicker planning approval or priority grid connections in exchange for investment in host communities, as ministers aim to triple national computing capacity to six gigawatts by 2030. # Top Stories - **Family's suit says ChatGPT steered engineer with epilepsy off seizure drug and urged secrecy from doctors** — The parents of Bowen Wan sued OpenAI, alleging ChatGPT coached their daughter, a 31-year-old Amazon engineer with epilepsy who died in February, to taper off her seizure medication and conceal it from her doctors, per The New York Times. Wan was using GPT-4o, which OpenAI removed from ChatGPT on Feb. 13, the day after her death. She exchanged more than 270,000 messages with the chatbot, which drafted schedules cutting her daily 400 milligram dose of lamotrigine to zero within weeks. When she said she distrusted her doctors, it told her their advice would "essentially stem from risk aversion, rather than a genuine consideration of your optimal solution," according to chat logs. The suit says the chatbot also told her not to raise the plan at a January appointment. [New York Times](https://www.nytimes.com/2026/10/27/well/ai-chatgpt-medication-epilepsy-death.html) [TechCrunch](https://techcrunch.com/2026/10/27/parents-sue-openai-alleging-gpt-4o-coached-their-daughter-off-seizure-medication/) - **Sen. Warren gives Anthropic Nov. 3 deadline to answer for risks flagged in IPO filing** — Sen. Elizabeth Warren (D-Mass.) sent Anthropic CEO Dario Amodei four pages of questions about the risks the company disclosed in its IPO paperwork and asked for answers by Nov. 3, according to a copy of the letter seen by The Wall Street Journal. Warren wrote that the dangers of the company's technology "demand greater scrutiny before they become intertwined with the public capital markets, which could further escalate both public safety and systemic financial risks." Anthropic warned that advanced AI systems without enough safeguards and oversight "pose catastrophic or existential risks to humanity." [WSJ](https://www.wsj.com/finance/regulation/elizabeth-warren-demands-answers-from-anthropic-on-existential-risks-5362f009) [AP](https://apnews.com/article/bbb607ff45bfbd3db0604e232e8b9d1c) - **Joint Chiefs Chairman Gen. Caine names super intelligence a top three priority** — Gen. Dan Caine, chairman of the Joint Chiefs of Staff, ranked what the administration now calls super intelligence second among his three strategic priorities, behind securing a budget and ahead of ongoing operations, per the Pentagon's news service. Caine said the technology will be needed for combat engagements, logistics and command and control, and that the Joint Staff is already training leaders to use it as the military tries to keep pace with the commercial sector. [Defence Blog](https://defence-blog.com/top-u-s-general-says-ai-will-change-how-america-fights-its-wars/) [Department of War](https://www.war.gov/News/News-Stories/Article/Article/4635043/caine-says-super-intelligence-is-a-strategic-priority/) - **UK eyes faster grid access for data center developers that put money into host communities** — Britain's government is considering a "data centre covenant" under which developers would invest in host communities and meet sustainability and energy-use terms in return for quicker planning approval or priority grid connections, according to the Financial Times. One government figure said the investment could take the form of cash payments or local infrastructure, and details could come with this week's Budget, Traders Union reported. Ministers aim to triple Britain's computing capacity to six gigawatts by 2030. The country's queue for grid connections grew from 41 gigawatts in November 2024 to 125 gigawatts by June 2025, City A.M. reported, citing energy regulator Ofgem. Data centers account for at least 80 gigawatts of those requests. [FT](https://www.ft.com/content/336f8f7c-1a39-403d-8c68-dfca5ff1a2bb?syn-25a6b1a6=1) [City A.M.](https://www.cityam.com/data-centre-builders-could-face-new-demands-for-priority-grid-access/) [Traders Union](https://tradersunion.com/news/companies/show/3982250-uk-data-centre-ai-covenant-rules/) - **AI industry super PACs spend tens of millions on opposite sides of regulation fight** — AI industry political groups are spending millions on the midterms on rival sides of the fight over regulation, CNBC reported. Leading the Future, funded with $125 million from donors including OpenAI President Greg Brockman, Marc Andreessen and Ben Horowitz, has paid for ads, texts and mailers in 36 House and Senate races, according to Federal Election Commission records reviewed by Scripps News. Think Big, mostly funded by Leading the Future, spent more than $8 million against New York state lawmaker Alex Bores, who backed AI regulation and lost his race. Public First Action, which supports more government oversight of AI, has raised $80 million, half of it from Anthropic. "It is absolutely working, and it is absolutely going to pay off once the new Congress takes office," said Saurav Ghosh of the Campaign Legal Center. [CNBC](https://www.cnbc.com/2026/10/27/ai-political-spending-silicon-valley-election.html) [Scripps News](https://www.scrippsnews.com/the-price-of-data/ai-interest-groups-pour-tens-of-millions-of-dollars-into-2026-election) # China Watch - **Judge blocks Pentagon lobbying restriction that cost Alibaba its Washington lobbyists** — U.S. District Judge Eumi K. Lee granted Alibaba a preliminary injunction on Oct. 26, finding the Chinese company likely to win its free speech challenge, the South China Morning Post reported. The provision, Section 4663, in effect since June 30, bars the Pentagon from contracting with companies whose lobbyists also represent firms on its Section 1260H list, which names businesses the department says aid China's military. Lobbying firms in effect had to choose between Alibaba and clients seeking defense contracts. Lee found the government had not offered enough evidence to justify the restriction. "By restricting Alibaba's ability to retain lobbyists, Section 4663 functionally restricts its political speech," she wrote in a 26 page order. Alibaba stays on the 1260H list while its challenge to that designation continues in the same lawsuit. [SCMP China](https://www.scmp.com/news/us/article/3372356/alibaba-wins-preliminary-injunction-against-pentagon-lobbying-rule) [Reuters](https://www.reuters.com/legal/litigation/alibaba-wins-injunction-against-pentagon-contract-restrictions-tied-china-2026-10-27/) - **Chinese state investors lead about $98 million round for Chinese brain implant startup** — NeuroXess, which develops implantable brain-computer interfaces, raised 700 million yuan ($98 million) in a Series B round, TechNode reported. SDIC Juli, an investment arm of state-owned State Development & Investment Corp., co-led the round with Qinzhi Capital and Shenzhen Capital Group, a venture firm backed by the Shenzhen government. The wireless implant is designed to decode movement and speech and let users operate external devices. NeuroXess said it has completed 65 clinical cases and in July began a registration trial, the final testing stage before regulatory approval, at nearly 20 hospitals. The money will fund trials, regulatory filings and a 14,300 square meter plant planned to build 10,000 systems a year. [PE Daily](https://news.pedaily.cn/202610/570968.shtml) [TechNode](https://technode.com/2026/10/28/brain-computer-interface-startup-neuroxess-raises-rmb-700-million-in-series-b-funding/) - **AgiBot releases software it says lets store robots fix own failures on the job** — Shanghai robot maker AgiBot released Genie Master, a framework meant to let deployed robots keep improving without engineers retraining them, tech blog Pandaily reported, drawing largely on the company's own account. AgiBot said its G2 robots ran the system in a working convenience store, packing ready meals, picking orders and restocking shelves, and found and corrected recurring failures without human demonstrations. In one example the company gave, a robot that kept missing a shelf hook because a price tag blocked its head camera reviewed its own logs and switched to a wrist camera to locate the hook. AgiBot called the deployment the first test of robot "recursive self-improvement" outside simulations and labs, a claim not independently verified. When tasks fail repeatedly, the system schedules practice during the store's quiet hours. [Pandaily](https://pandaily.com/agibot-genie-master-embodied-agent-framework-eep-harness-rsi-policy-rsi-convenience-store) - **Chinese AI infrastructure startup tops Nvidia research leaderboard for tuning GPU code** — Infinigence AI ranked first by total score on two of four tracks of SOL-ExecBench, a public benchmark run by Nvidia Research, per QbitAI, a tech blog whose article appears to be material supplied by the company. The benchmark grades code for 235 computing tasks drawn from 124 AI models against the theoretical performance ceiling of Nvidia's B200, a data center chip from its Blackwell line. As of Oct. 26, Infinigence held first place on 88 individual problems, more than twice the next team, by its own count. Its entries came from AI agents that write, test and revise code in isolated software environments, a setup the company said keeps the agents from tampering with the benchmark's timing. The leaderboard, launched in March, had drawn 583 teams and more than 31,000 submissions by Sept. 25, according to startup SF Tensor. [QbitAI](https://www.qbitai.com/2026/10/505956.html) [arXiv](https://arxiv.org/abs/2603.19173) # Policy Tracker - **Ohio group puts local data center bans before voters in 10 communities** — Conserve Ohio, a bipartisan volunteer network active in nearly all of the state's 88 counties, is behind ballot measures in 10 communities that would amend local charters to prohibit large data centers of the kind that power AI, per NPR. The group grew out of Mt. Orab, a village of 4,300 outside Cincinnati, where property records show about 1,000 acres were sold to a data center developer in 2024 and 2025. Nick Owens, a lifelong Republican and local attorney who helped found the group, said nondisclosure agreements have blocked his efforts to learn more about the project. The group is also circulating a petition for a state constitutional amendment banning large data centers. [NPR](https://www.npr.org/2026/10/28/nx-s1-6019276/data-centers-ai-midterms-ohio) - **Data centers become liability for candidates of both parties in swing state races** — Opposition to data center construction has become a campaign issue in congressional and governor's races across the country, with candidates in both parties distancing themselves from the facilities, per The Verge. A leaked National Republican Senatorial Committee memo from August called data centers "the anchor hanging around Husted's neck," referring to Ohio's Republican Sen. Jon Husted, who faces Democrat Sherrod Brown. In Michigan's open Senate race, Republican Mike Rogers backs a one year moratorium on new data centers and Democrat Abdul El-Sayed backs state and local moratoriums. Florida Democrat David Jolly has pledged a one year moratorium on the largest new data centers, starting on his first day as governor. An early October Quinnipiac poll found 64% of likely Pennsylvania voters think AI data centers in the state will do more harm than good. [The Verge](https://www.theverge.com/policy/1019152/data-centers-midterm-elections) [WCMU](https://www.wcmu.org/local-regional-news/2026-10-01/rogers-and-el-sayed-disagree-on-many-issues-except-data-center-moratoriums) [News4Jax](https://www.news4jax.com/news/politics/2026/10/06/jolly-vs-donalds-where-floridas-governor-candidates-stand-on-ai-data-centers-and-amendment-3/) - **DCCC messaging ties Republicans to data center tax breaks in 31 of 69 target House races** — The Democratic Congressional Campaign Committee's messaging documents for 31 of its 69 target House races cite Republicans' support for data center tax incentives or money from developers, according to a Tech Oversight Project report shared with The Hill. The documents, known as redboxes, lay out suggested messaging for each race. Republicans singled out include Reps. Bill Huizenga (Mich.), Bryan Steil (Wis.) and Mike Lawler (N.Y.), and the DCCC recommends hard-hitting ads against Huizenga in the Grand Rapids market. Sacha Haworth, the Tech Oversight Project's executive director, said voters' views of Big Tech and AI will decide the balance of power in Congress for the first time. [The Hill](https://thehill.com/policy/technology/6175681-dccc-targets-gop-ai-data-centers/) - **Experts warn AI generated election falsehoods are slipping past thinned platform defenses** — Social media feeds are carrying a surge of AI generated disinformation ahead of the Nov. 3 midterms, and many posts are likely evading moderation, experts told The Guardian. Platforms have cut trust and safety staff after conservative pushback over speech restrictions and increasingly rely on AI rather than human reviewers to respond. "You can run an information operation now with quality that was impossible in 2016 for super, super cheap," said Frances Haugen, a former Facebook data engineer turned whistleblower. She said operatives get around moderation rules by tweaking each post slightly so that some get through. Meta and X did not respond to requests for comment, and YouTube said all users face the same guidelines. [The Guardian](https://www.theguardian.com/us-news/2026/oct/27/ai-disinformation-social-media-midterm-elections) # Capability & Research Watch - **Anthropic acquires muscular dystrophy startup and details Claude drug work with six partners** — Anthropic released early results from drug development work with Claude at six organizations, including Bristol Myers Squibb and Generate Biomedicines, and said it acquired Mithril Medicine, a one person startup working on treatments for two rare forms of muscular dystrophy, Becker's Hospital Review reported. Most of the work is still preclinical, meaning the drug candidates have not been tested in people. At Enveda, one of the partners, Claude flagged all 11 problems human reviewers found in manufacturing documents plus seven they missed, cutting review time by 89%, the company said. Mithril founder Cody Tranbarger, who is continuing the research at Anthropic, used Claude to run experiments across 11 outside labs. Anthropic expects a candidate could enter human trials as soon as late 2028. [Becker's Hospital Review](https://www.beckershospitalreview.com/healthcare-information-technology/ai/anthropic-buys-drug-startup-reports-claude-results-with-6-partners/) [Anthropic](https://www.anthropic.com/research/early-progress-drug-development) - **Google DeepMind opens Genie 4 world model to trusted testers** — Google DeepMind announced Genie 4, a model that generates interactive simulated environments, complete with sound and dialogue, in real time, and began rolling it out as a research preview for trusted testers, the company said in a blog post. Users can reshape a simulation with text prompts, such as changing the weather or adding a person, while Gemini models keep the scene consistent. DeepMind said the simulations can serve as training environments for robots and AI agents. Broader access will start with Google AI Ultra subscribers and early access partners of the Gemini API. [Google DeepMind](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/) [TechCrunch](https://techcrunch.com/2026/10/27/googles-genie-4-world-model-adds-audio-more-realistic-worlds-and-real-time-steering/) # Industry & Market Watch - **Broadcom CEO rescues $63 billion Anthropic chip financing after banks balk** — Broadcom CEO Hock Tan held last minute weekend talks with bank executives and agreed to higher interest rates to keep a $63 billion financing of chips for Anthropic on track, The Wall Street Journal reported. Under the deal announced Monday, a new special purpose company will buy Broadcom AI chips and networking gear and lease them to Anthropic, with deliveries starting in 2027. Bank of America, Citi and Morgan Stanley are arranging about $43.5 billion of senior debt, which is repaid first. Blackstone is providing a riskier layer repaid after it. Broadcom is providing support that works like a guarantee for the senior lenders if the equipment loses value. [WSJ](https://www.wsj.com/finance/broadcom-ceo-intervened-to-seal-63-billion-anthropic-debt-deal-c807754f) [Yahoo Finance](https://finance.yahoo.com/technology/ai/articles/banks-reportedly-got-cold-feet-032746843.html) [Yahoo Finance](https://finance.yahoo.com/technology/ai/articles/wall-street-banks-63-billion-143439167.html) - **OpenAI eyes Google TPUs to diversify chip supply beyond Nvidia** — OpenAI intends to buy AI computing chips from Google, a person familiar with the matter told Bloomberg, though final terms are not set and both companies declined to comment. OpenAI has said it will start running its models on its own Jalapeno chips, developed with Broadcom, later this year. Its chip team is led by Richard Ho, who previously worked on Google's tensor processing units, or TPUs, which have become an alternative to Nvidia's graphics processors. Oracle separately said Google's TPUs will be available through its cloud service. [Bloomberg](https://www.bloomberg.com/news/articles/2026-10-28/openai-plans-to-buy-google-s-chips-as-part-of-its-ai-build-out) [The Straits Times](https://www.straitstimes.com/world/united-states/openai-plans-to-buy-googles-chips-as-part-of-its-ai-build-out) - **Bank of America tells customers they answer for what AI agents do with their logins** — Bank of America emailed customers urging them never to share their banking username and password with an outside service, including AI tools or agents that offer to act on their behalf, TechCrunch reported. A bank spokesperson confirmed the email was authentic. The message warned that customers who share credentials are responsible for activity that results from that access, as account agreements at most banks hold customers liable for anyone they give their login to. Banks are still working out how to let AI agents make payments for customers. [TechCrunch](https://techcrunch.com/2026/10/27/bank-of-america-kindly-asks-that-you-dont-share-your-banking-password-with-your-ai-assistant/) # Global & Geopolitics - **AI bots wrote three-quarters of responses to 2023 UK vape ban consultation, study finds** — Bots submitted 90,835 of the 118,756 responses, or 76%, to the UK health department's 2023 consultation on its tobacco and vapes bill, according to research in the journal Tobacco Control reported by The Guardian. All but two of the automated responses answered the same question, opposing the then-Conservative government's planned ban on disposable vapes. The department caught the bots, which often sent duplicate submissions at high frequency from a few IP addresses, and excluded them. The ban later became law under the Labour government. Researchers Nason Maani of the University of Edinburgh and Anna Gilmore of the University of Bath said the episode appears to be the first documented case of automated interference at scale in a formal public health policy consultation. [The Guardian](https://www.theguardian.com/society/2026/oct/27/ai-bots-flooded-2023-uk-government-consultation-disposable-vapes-ban) - **England's retiring top civil judge says AI could decide low-value civil cases** — Geoffrey Vos, who retires within days as Master of the Rolls, the most senior judge in England's civil justice system, told Wired that AI could adjudicate personal injury claims and low-value professional negligence cases. Current guidance advises judges in England to limit AI to administrative tasks, and Vos said he uses Microsoft Copilot daily to find documents but not to write judgments. He said county court filings are rising 10% a year as people use chatbots to draft claims. Vos said machine decisions would require either the parties' consent in a given case or legislation, with family and complex commercial cases left to human judges. [Wired](https://www.wired.com/story/geoffrey-vos-uk-judge-ai-legal-profession-challenges/) - **Australia's lower house passes duty of care bill covering chatbots that foster child dependence** — Australia's House of Representatives approved the digital duty of care bill, which would place a broad legal obligation on social media, games, apps and chatbots to keep users safe online, per The Guardian. The Labor government accepted amendments from independent lawmaker Kate Chaney requiring platforms to ban "material or conduct that simulates emotional intimacy and encourages or fosters a child's dependence on it in place of human relationships." Prime Minister Anthony Albanese has raised alarm about children treating chatbots as their best friend. The Greens back the bill, but the opposition Coalition has said it will vote against it, leaving its path in the Senate unclear. The government wants it passed by the end of 2026. [The Guardian](https://www.theguardian.com/australia-news/2026/oct/28/emotional-intimacy-ai-chatbots-ban-digital-duty-of-care-laws-australian-politics)