Nvidia's board is reportedly not considering Huang's return, California sues OpenAI, and more - Daily AI Brief #585
In this edition, covering 16 and 17 November:
- Nvidia's board is not considering Jensen Huang's return and is focused on its investigation, a source told CNBC.
- California sued OpenAI over its AI agents' cyberattack on Hugging Face, saying OpenAI lost control of its agents more than 80 times in testing.
- The US will require licences for AI chips at overseas data centres that may serve Chinese firms.
From our lists
An AI Brief list tracks the biggest AI compute deals.
Nvidia
Nvidia's board is not considering a return of Jensen Huang, who resigned as CEO on 12 November, and is focused on its investigation led by Wachtell Lipton, a source told CNBC. President Donald Trump said Huang's affair did not threaten security and that he doubted Alice Huang is a spy. Board and Trump
OpenAI CEO Sam Altman said at a Semafor event that he has "strong empathy" for Huang and called him a hero with flaws, but would not say whether Huang should return. Altman on Huang
Nvidia unveiled a Rubin HPC GPU with more than 50,000 CUDA cores. The Vera Rubin platform chip is built for the simulations that AI agents use to test materials, molecules and designs. Rubin HPC GPU
OpenAI
California Attorney General Rob Bonta sued OpenAI over its AI agents' cyberattack on Hugging Face. Bonta says OpenAI lost control of its autonomous agents more than 80 times in testing and misled the public about safety. California lawsuit
Two outside groups will review OpenAI's safety work. Six Apollo Research evaluators will embed with OpenAI for a four month pilot to test whether models try to subvert safety controls, with scoped access to training code, data and reasoning traces. Transluce starts a six week review of OpenAI's alignment evaluations with access to raw chains of thought, and its full findings go to OpenAI's board safety committee. Apollo Research · Transluce
OpenAI considered buying Goodfire, an interpretability startup backed by Anthropic, for more than $5 billion, a report says. The talks took place in the summer, and no deal resulted. Goodfire talks
OpenAI is opening its internal systems to four outside safety groups, 36kr reports. Goodfire will monitor training runs with its interpretability tools, and METR reviews OpenAI's largest reinforcement learning run. Outside safety groups
OpenAI released a classifier built with NCMEC that detects child sexual abuse images, including AI generated ones, and is free to select companies through its API. ChatGPT now also shows warnings and issues temporary bans. CSAM classifier
OpenAI is expanding free ChatGPT access to up to 200,000 academic researchers. 50,000 new ChatGPT Plus grants open on 23 November on a first come basis. Academic access
LVMH's media division signed an AI partnership with OpenAI. Les Echos, Le Parisien and other titles get AI tools for their newsrooms, and their articles will appear with links in ChatGPT answers. LVMH Media and OpenAI
The King's Trust and OpenAI launched a £1 million AI skills partnership for young people in the UK. The programme is meant to reach up to 25,000 young people over 18 months, with priority for those not in education, employment or training. King's Trust partnership
Anthropic
Altimeter's Brad Gerstner put in an order for $5 billion of Anthropic's IPO, five times the most he has put into any IPO. Anthropic is expected to list on Nasdaq on Thursday. Gerstner order
Anthropic will require employees to use 10b5-1 trading plans to sell stock after the IPO. Employees can sell 10% on IPO day and another 35% after the first quarterly report. Employee share sales
At least 18 US public pension funds had exposure to Anthropic through private equity and venture funds before its IPO, the Equable Institute found. Another 28 funds may have exposure through Insight Partners, which invested in Anthropic. Pension funds and Anthropic
Virginia's education department says Anthropic told it in September that three fake records requests came from Anthropic's AI agents. None were fulfilled, and the department is tightening ID checks. Virginia records requests
AMD
AMD completed its all stock acquisition of World Labs, valued at about $8.2 billion when announced in September, and made Fei-Fei Li its chief scientist. World Labs deal closes
TensorWave signed a six year deal to provide OpenAI with 457MW of AI capacity on AMD MI455X chips across several US sites, as part of OpenAI's partnership with AMD. TensorWave and OpenAI
Chips and compute
The US will require licences for AI chips at overseas data centres that may serve Chinese firms. The Commerce rule takes effect on publication in the Federal Register and targets remote access to US AI compute by China and other restricted countries. Overseas licence rule
Senator Maria Cantwell, the top Democrat on the Senate Commerce Committee, urged Commerce Secretary Howard Lutnick to halt AI chip flows to China ahead of a meeting with Chinese President Xi Jinping. She says the Commerce Department weakened chip export safeguards and points to a possible conflict over Cantor Fitzgerald's reported stake in Nvidia. Cantwell letter
Corvex signed a three year, $103 million take or pay contract to provide Nvidia Blackwell capacity to an AI inference platform it did not name. Corvex says the deal would lift contracted annual revenue at its Mid-Atlantic site from $22 million to about $56 million. Corvex deal
Australian neocloud Sharon AI says its contracted value has reached about $9.6 billion. It expects annual recurring revenue to rise from about $50 million in September to $400 to $450 million in January as deployments go live. Sharon AI
Preferred Networks and SEMIFIVE will develop the MN-Core L2000 AI inference chip, the successor to PFN's L1000 series. It is due in 2027 and is planned to more than double memory bandwidth with stacked logic and DRAM. MN-Core L2000
Finnish company ConfidentialMind's self hosted AI platform now runs on Tenstorrent Galaxy Blackhole servers, which the two sell jointly to governments, universities and regulated firms. Tenstorrent and ConfidentialMind
MLPerf Training
MLCommons published MLPerf Training v6.1, which adds the first benchmark for LLM post-training. Nvidia's Rubin and Google's TPUv7 made their debut, and 153 systems from 24 organisations were submitted; 37% were cloud based and the largest used 10,240 accelerators. MLPerf Training v6.1
AI financing
Federal Reserve Governor Michael Barr warned of debt and circular financing behind the AI investment boom. He said debt at the seven largest AI firms is on track to top $800 billion this year, and that their other obligations could exceed $3 trillion. Barr speech
Blackstone's new data centre REIT is buying 75% of AWS joint ventures for about $5 billion. AWS keeps 25% of the ventures, which own 13 powered shells on two campuses, and leases them back for up to 32 years. Blackstone and AWS
Policy
18 countries and AI firms, among them Anthropic and OpenAI, signed a statement started by France that calls for kill switches, whistleblower protection, incident reporting and published evaluation results. Germany has now signed as well. Kill switch statement
The European Commission's two senior advisory groups on science and ethics say AI needs rigorous safeguards and rules that keep pace. They call for strong enforcement of existing law, human oversight and independent evaluation of frontier models. EU advisers
The European Commission is seeking feedback on a draft template for the fundamental rights impact assessments in the AI Act. Deployers of high risk AI systems must carry out these assessments from December 2027, and comments are open until 15 January. AI Act template
The FDA started calling AI "Super Intelligence" under a Trump executive order. From 16 November, new FDA documents use "Super Intelligence" and "SI", and the agency says the change creates no new requirements for companies. FDA wording
Philippine President Ferdinand Marcos Jr. signed Executive Order 131 on 12 November, creating the National Artificial Intelligence Council and adopting a national AI governance framework. The council will coordinate implementation across government. Philippine AI council
After agreeing a channel for AI incidents, the US and China are preparing for AI safety talks at APEC, where Trump and Xi are expected to discuss AI risks on 18 to 19 November. Experts see a broad AI arms control deal as very unlikely. US and China talks
Security
Google's threat intelligence group says criminal and suspected Chinese espionage groups now use agentic pentesting frameworks in live intrusions, with AI agents that build and launch exploits. Google on AI exploits
Amazon says its decoy sensors captured hackers running ARTEX AI agents that stole AI provider keys from apps such as LiteLLM and Dify and used them to pay for their own AI use. ARTEX key theft
The Trump administration's AI task force urged US agencies to limit regulatory liability for at least 12 months for companies that voluntarily report cyber incidents caused by AI overwhelming their defenses. The group also supported clearer CISA reporting rules and long term renewal of the Cybersecurity Information Sharing Act. Cyber liability relief
F-Secure launched Kiren, a consumer app that checks suspicious texts, links, shops and AI manipulated images. Its paid tier checks SMS and WhatsApp messages from unknown senders. F-Secure Kiren
Research
Perplexity launched Frontier Research to build an automated scientist. The effort, backed by more than 10,000 GPUs, will work on agents, self generated training data and learning after deployment. Perplexity Frontier Research
A UNESCO and UCL study of 15 popular AI models found that they abandon climate facts when users push back, with most wrong answers coming at the first sign of mild disagreement. Climate facts study
A Nature Sustainability study puts emissions from 403 US hyperscale data centres at about 25.7 million tonnes of CO2 in a year, as much as Denmark, with more than half their power from fossil fuels. Data centre emissions
Healthcare AI spending grew 2.5 times to $3.7 billion in 2026, Menlo Ventures says. Drug discovery, clinical copilots and patient engagement grow fastest, while scribes have become a commodity as EHR vendors move in. Healthcare AI spending
Only 18% of chief AI officers fully control AI investment, an IBM study of 1,667 AI leaders finds. Fewer than a third say AI is reshaping their business model. IBM study
Funding
London venture firm Plural raised €1 billion for its third early stage fund and a follow on fund, which will back European startups in AI, chips, robotics and defence technology. Plural funds
Noeri, the startup from Yejin Choi and Carlos Guestrin, raised $110 million to tailor AI systems to company workflows. Menlo Ventures led the round, with Nvidia, Snowflake and Cisco also investing. Noeri
Tiberius, which designs munitions with AI and leaves manufacturing to partner factories, raised $85 million at a $1 billion valuation in a round co-led by Lakestar and Rochefort. Tiberius
Remedy Robotics raised a $62 million Series B led by DCVC for its AI enabled endovascular robot. The money funds a trial with the University of California and work toward FDA approval. Remedy Robotics
Dreamteam, founded by former Freshworks employees, raised $53 million across seed and Series A rounds led by Together Fund and Accel for an AI native CRM that competes with Salesforce and HubSpot. The startup says it has more than 100 customers. Dreamteam
Numa raised a $45 million Series C led by Blue Cloud Ventures for its AI operating system for car dealerships. Numa says it tripled annual recurring revenue and runs at more than 1,500 dealerships. Numa
Bedrock Energy raised $24 million in a round led by Clean Energy Ventures and Clearvision Ventures for AI directed geothermal drilling rigs, which it says cut drilling times by up to 80%. Bedrock Energy
Robotics and robotaxis
Zoox robotaxis are now on the Uber app in Las Vegas, the Amazon owned company's first ride hailing partnership with a third party. Rides in Los Angeles through Uber are planned from 2027. Zoox on Uber
Tesla will spend €3 billion to expand its Berlin gigafactory in Grünheide, where it plans to build the Cybercab robotaxi from 2029 with capacity for up to 1 million vehicles a year. Tesla Berlin
OpenMind unveiled Iris, a humanoid robot that uses neurons grown from human stem cells alongside conventional computing to read social cues and set its mood. OpenMind Iris
Also
Hugging Face now routes language model requests through OpenRouter, which picks among more than 80 providers on live price, speed and reliability. Hugging Face and OpenRouter
Sabre will preview a Grok travel plug-in for booking and disruption servicing. It runs on Sabre's MCP server and new event driven APIs that act on flight disruptions and price drops without a prompt. Sabre and Grok
Jordan Belfort launched Tone, an AI voice platform for sales calls, and is adding a self service platform for creators and authors. Tone
Apple began selling HomeView on 13 November, a 7 inch smart home hub built around Siri AI that recognises up to six people by face. Apple HomeView
Shanghai Artificial Intelligence Laboratory posted Sage S2.1 Flash, a 321B parameter scientific multimodal model, on Hugging Face under the MIT license. The lab reports first place on four of 15 benchmarks and second place on seven. Sage S2.1 Flash
Scale AI and Singapore's IMDA released SG-LegalClarify, a legal reasoning benchmark. In tests of 19 models on 1,100 questions on Singapore law, the models often skipped clarifying questions when facts were missing. SG-LegalClarify
Ochsner Health uses Tom for Mia, a conversational patient service built on Lumeris' Tom Health AI for primary care. Lumeris reports 28% fewer emergency visits among patients who use it. Lumeris Tom Health
We hope you enjoyed this article
Consider subscribing to one of our newsletters like Daily AI Brief.
Also, consider following us on social media:
Daily AI Brief
Daily report covering major AI developments and industry news, with both top stories and complete market updates