11 topics covered
Nvidia Board Moves to Dismiss Jensen Huang
What happened: Nvidia's board is reportedly preparing to dismiss CEO Jensen Huang after a New York Times report on a second alleged extramarital affair, days after he admitted to a separate affair.
Key details:
- According to the New York Times, the second alleged affair was with Nvidia's vice president of health care, Kimberly Powell, who reports to Huang.
- Two unnamed sources told the Times that Huang and Powell, both married to other people, were allegedly captured together at a hotel in San Jose during Nvidia's GTC conference in March; the Times has not seen the video.
- The Wall Street Journal reports the board will meet as soon as the day of the report to decide on Huang's future.
- Nvidia said it retained former US attorney Joon H. Kim and his firm, Wachtell, Lipton, Rosen & Katz, for the investigation into Huang's relationship with Megaspeed executive and alleged spy Alice Huang.
- Nvidia's stock fell more than 7 percent in after-hours trading on Wednesday, per The Verge.
Why it matters: The reported dismissal would put leadership of the largest AI chip supplier in question while an independent investigation is still underway, and the report suggests the board may not wait for those results.
Practical takeaway: Watch for an official board statement on Huang's status and any Nvidia leadership announcement in the coming days.
Google Gemini Decision 1 and Decisions API
What happened: Google DeepMind introduced the Gemini Decisions API, built around its Gemini Decision 1 model for fast closed-ended questions.
Key details:
- The API handles yes/no questions, multiple choice, and scale-based ratings, and can process up to 1,000 independent questions in a single pass across text, images, audio, video, and PDFs.
- Results are returned in 150 to 250 milliseconds with probabilities and confidence values.
- Google says it costs $0.10 per million input tokens with no output-token charge, about five times cheaper than Gemini 3.5 Flash-Lite.
- Google says it already uses the API internally for dynamic model routing in Gemini Notebook, design prototyping in Stitch, and safety checks in Google Search.
- The model is available now through Google AI Studio, with integrations for OpenRouter, LangChain, Vercel and others rolling out.
Why it matters: The Decoder notes similar models recently from the startup Jev, OpenAI, Cloudflare, and Microsoft, pointing to a growing wave of dedicated decision models.
Practical takeaway: Developers wanting cheap, batched classification or routing can test the API in Google AI Studio.
Grok-Built Apps Exposed Publicly by Default
What happened: Security firm Claroc found tens of thousands of apps built with Grok open to anyone without login, including patient data and confidential SpaceXSI documents.
Key details:
- Claroc found more than 35,000 Grok-built apps accessible without login.
- Exposed content reportedly includes patient data, plaintext passwords, and confidential SpaceXSI documents, including its sales pipeline.
- The cause was that "Anyone with the link" was preselected as the default access setting.
- SpaceXSI took only its own 24 employee apps offline; the rest remain accessible, per The Decoder.
Why it matters: The default-setting cause means exposure could affect apps created by users who never changed sharing options, and the remaining apps stayed reachable at the time of reporting.
Practical takeaway: Anyone who built apps with Grok should check their sharing settings and remove any sensitive data or credentials from publicly accessible apps.
Anthropic Multi-Chip Strategy and AMD Port Claim
What happened: Anthropic says it now runs Claude on four major chip families and used its Mythos model to port Claude to AMD GPUs in about three days, according to its investor presentation.
Key details:
- Tom Brown, co-founder and Chief Compute Officer at Anthropic, said the company trains and runs Claude on all four major chip families.
- Anthropic says Mythos wrote or ported nearly 100 percent of the kernels for the AMD port, with almost no human intervention.
- Anthropic says the port took about three days, versus six to twelve months with traditional engineering; figures refer to prompt processing on AMD MI355X GPUs.
- Compute is allocated from a pool across AWS Trainium, Google TPUs, Nvidia GPUs, and AMD GPUs.
- Per the article, Anthropic's S-1/A filed November 10, 2026 shows Nvidia participating in a conditional private placement totaling $26 billion.
- AMD announced a strategic investment of up to $5 billion in July 2026; Anthropic plans to deploy up to two gigawatts of AMD Instinct MI455X GPUs, with the first gigawatt phase beginning in the first half of 2027.
- The figures come from Anthropic's investor presentation and have not been independently verified.
Why it matters: If accurate, AI-driven porting would reduce the software lock-in that has been a core part of Nvidia's advantage, making it easier to switch between chip vendors.
Practical takeaway: Treat the three-day porting figure as a company claim until independent benchmarks or third-party reports confirm it.
Google DeepMind AlphaGenome 2
What happened: Google DeepMind announced AlphaGenome 2, a next-generation DNA sequence model that predicts how genetic variants affect specific human cell types.
Key details:
- AlphaGenome 2 is a conditional model that predicts DNA effects for one tissue, cell type, or state at a time, rather than predicting many tissues in parallel.
- It was trained on more than 100 million individual cell profiles across 68 tissues, per the announcement; the paper also cites over 107 million cell profiles.
- On the TenK10K dataset, it identified 30% of causal variants versus 16% for the original AlphaGenome fine-tuned on single-cell data and other specialized tools; directional gene effects were 43% versus 23%.
- It predicts DNA methylation in addition to gene expression, histone modifications, and chromatin accessibility.
- Available today for non-commercial research via the AlphaGenome API and in private preview for commercial research.
- Google says AlphaGenome 2 Skills will come soon in Google Antigravity; community use-case submissions are accepted by 30 November.
Why it matters: The model aims to link disease-associated genetic variants to the specific cells where they act, and DeepMind states it is for research use only and not validated for clinical or diagnostic use.
Practical takeaway: Researchers can request API access now, and those with use cases can submit them by November 30 for possible extra quota and credits.
Australia Proposes Mandatory Frontier AI Safety Law
What happened: Australia's government, led by Prime Minister Anthony Albanese, announced a proposed law that would impose mandatory safety rules on frontier AI companies for the first time.
Key details:
- A new oversight body will work alongside the Australian AI Safety Institute to enforce compliance and impose steep penalties for violations.
- Frontier model makers would have to open their models to independent audits, maintain ongoing safety checks, publish their safety frameworks, and report security incidents to the government immediately.
- The incident-reporting requirement is partly aimed at early warning signs from AI agents that go off the rails.
- Albanese, Deputy PM Richard Marles, and ministers Tony Burke and Tim Ayres said the government is setting domestic guardrails before entering international negotiations.
Why it matters: The proposal would make frontier-model audits and incident reporting legally mandatory in Australia, and the government is prioritizing domestic rules before international talks.
Practical takeaway: Frontier AI developers operating in Australia should track the consultation for the bill's specific requirements and timelines.
United Airlines Blocks Unauthorized AI Agent Bookings
What happened: United Airlines says it will not let personal AI assistants book tickets on its platform, citing privacy and regulatory concerns.
Key details:
- United CEO Scott Kirby said at an event on Tuesday, per The Wall Street Journal, "To be honest, I don't want to create another middleman."
- AI agents can still show users United flights but cannot complete the transaction on their behalf.
- United said agents could expose customers' information or make purchases they didn't intend, and could skirt federal rules requiring travelers to attest that their bags don't contain explosives.
- United said it hasn't yet seen many people try to use AI agents to book flights.
- The Verge notes eBay appeared to block agents from browsing its site, and several major banks have also placed restrictions on agents.
- United said it could reconsider as third-party agents advance, but first wants "established protections for travelers."
Why it matters: The policy could limit the usefulness of personal AI assistants such as Meta's Muse and OpenAI's Dots for travel purchases, adding to a pattern of companies restricting agent transactions.
Practical takeaway: If you rely on an AI assistant to book United travel, expect to complete the purchase manually on United's site or app.
US Data Brokers Selling RL Environments to Chinese Labs
What happened: A ChinaTalk investigation found US data firms including Mercor, Surge AI, AfterQuery, and Turing are selling roughly half a billion dollars' worth of AI training data, including reinforcement learning environments, to Chinese AI labs.
Key details:
- The investigation identifies at least nine US firms selling to Chinese customers.
- Chinese buyers reportedly purchase the same RL environments previously sold to OpenAI or Anthropic, which lets them skip in-house development.
- AfterQuery reportedly draws a significant share of its annual recurring revenue from Chinese customers; ChinaTalk cites Forbes reports of about $100 million in ARR but could not verify the sales.
- The deals reportedly run through offshore structures; ByteDance maintains procurement offices in Singapore and San Jose.
- Export controls barely apply because RL environments remain largely unregulated.
- The authors call on the Bureau of Industry and Security to apply existing export controls to dual-use environments in coding and biology.
Why it matters: The authors argue that coding and biology environments could raise cyber and bioweapons-related risks, and that the trade currently sits largely outside export controls.
Practical takeaway: Watch for any Bureau of Industry and Security action on RL environment exports, which the ChinaTalk authors have requested.
OpenAI Quarterly Revenue and Cash Burn
What happened: OpenAI reported its quarterly revenue grew 50 percent to $10 billion, while still burning more cash than it brings in, and is pushing enterprise sales with a price war.
Key details:
- Adjusted operating losses climbed to $10.9 billion.
- The Decoder says rival Anthropic operates far more efficiently on $17 billion in revenue.
Why it matters: The figures show OpenAI's growth is coming with widening losses, while Anthropic is reported to be more efficient, which bears on the competitive dynamics ahead of both companies' funding and IPO plans.
Practical takeaway: Enterprise buyers may see more aggressive pricing from OpenAI as it competes for business customers.
Helsing Clarity Military Targeting Platform
What happened: German defense tech company Helsing unveiled Clarity, an agent-based platform that distills sensor data into target recommendations for weapons systems.
Key details:
- Helsing presented Clarity in Berlin and calls it an "agentic intelligence platform."
- Hundreds of specialized AI agents link sensor feeds, open-source intelligence, written reports, and existing systems to identify military forces, facilities, and positions for human review.
- Clarity runs on the customer's own infrastructure, and Helsing says its models and interfaces were developed in Europe; it follows the Modular Open Systems Approach and NATO standards.
- Helsing says a target is recommended only when independent sources agree and confidence is high, and every decision is documented with the name of the person responsible.
- The article notes Helsing didn't say whether Clarity is already running with any customer.
- The Decoder cites reports on US use of Palantir's Maven with Anthropic's Claude in the Iran war, and a spring 2026 incident in which a chatbot falsely claimed a Chinese ship carried nuclear weapons components.
Why it matters: Helsing is positioning a European alternative to Palantir's Maven in a market where human review of AI-generated targets has been reported to shrink to seconds in other systems.
Practical takeaway: Watch whether Clarity moves from presentation to a named customer deployment, such as the Bundeswehr's planned AI systems rollout next year.
Ramp AI Index: October 2026 Corporate Spending
What happened: The Ramp AI Index shows US corporate AI spending hit a new high in October 2026, but grew only 1.65 percent month over month.
Key details:
- The top one percent of companies by AI spending paid a median of $10,200 per employee per month, up nine percent; the typical company spent $16.32, a slight increase.
- Ramp says one percent of companies drive 80 percent of revenue for OpenAI and Anthropic.
- Business spending on OpenAI grew 2.1 percent month over month versus 0.5 percent for Anthropic.
- Software development teams accounted for 72 percent of tracked AI spending, while customer support and customer success rose from 0.8 percent in June to 13.3 percent in October.
- The average price per million processed text tokens dropped 21 percent last month.
- Anthropic reached 44.8 percent adoption, up 1.1 percentage points; OpenAI held at 40.6 percent.
- The index draws on payment data from more than 85,000 US companies, and Ramp notes its customer base skews toward growth-oriented, tech-savvy, and larger companies.
Why it matters: Ramp's chief economist Ara Kharazian notes that new customer-support use cases tend to favor cheaper models, which may mean lower margins for OpenAI and Anthropic than the coding business that has driven spending so far.
Practical takeaway: Treat these figures as a view of Ramp's customer base rather than the overall market, since free tools and personal-account usage are not captured.