AI (artificial intelligence): analysis of 12 November 2026
Back to the Chronology · Current news in this section

Analysis
Nvidia's board is reportedly preparing to dismiss Jensen Huang after a second alleged affair, days after he admitted to a relationship with a woman alleged to be a Chinese spy, in a scandal framed as a national-security matter amid the US–China AI infrastructure race.
The dominant story is leadership turmoil at Nvidia, where the Wall Street Journal reports the board is preparing to fire CEO Jensen Huang following a New York Times report of a second alleged extramarital affair; none of this is confirmed by Nvidia and the allegations remain reported rather than established 28. Alongside this, Google entered the emerging 'decision model' category with Gemini Decision 1, a cheap, fast API for scoring up to 1,000 closed-ended questions, closely mirroring an existing product called Jev 11220. AWS released a free local emulator of more than 200 services whose code was written and is maintained entirely by AI agents without engineer review, a notable milestone for agentic software production 1329. Beneath the product news, financial signals are mixed: CoreWeave is leaning on prepayments and shorter contracts, corporate AI spending growth has slowed sharply, and Microsoft researchers disclosed a backdoor technique targeting tool-using agents 242733.
Key developments
- Nvidia leadership crisis. The Verge, citing the Wall Street Journal, reports Nvidia's board is preparing to dismiss Jensen Huang after the New York Times reported a second alleged affair with an executive, days after Huang admitted to a separate affair with a woman alleged to be a Chinese spy [8]. TechCrunch frames the affair as having potential national-security ramifications given the US–China race in AI infrastructure [2]. The board's intentions and the second allegation are unconfirmed reports; any change at the top of the dominant AI chip supplier would reverberate across the sector, including customers and partners such as CoreWeave [24]. 2824
- Google enters the decision-model category. Google launched the Gemini Decisions API with Gemini Decision 1, returning calibrated probabilities for up to 1,000 yes/no, choice and rubric-scoring questions across text, images, audio, video and PDFs at $0.10 per million input tokens, with latency of 150–250 milliseconds and roughly five times lower cost than Gemini 3.5 Flash-Lite according to Google [12][20]. Simon Willison notes the API is almost identical to Jev's, down to reusing its 'Noul' term for yes/no questions and matching its pricing, with OpenAI also already in the category [1]. This signals a fast-commoditising niche for cheap, high-volume classification and grading tasks, with Google following rather than defining the design 11220
- AWS ships an emulator written entirely by AI agents. AWS released a free AWS Emulator simulating more than 200 services locally, saying AI agents wrote all of its code without review by its engineers and continue to maintain it by testing against the real cloud [29]. Willison reports it is written in Rust with 29,684 commits in the project's first three weeks [13]. It is one of the most prominent examples yet of a major vendor shipping agent-generated software at scale, and sits alongside AWS's broader push on agent tooling such as the AgentCore Web Search API and automated agent registry updates [6][15]. 1329615
- Strain and slowdown in AI economics. CoreWeave's third-quarter earnings show it increasingly asking customers to pay upfront, borrowing against specific contracts and considering shorter-term deals that could raise prices [24]. The Ramp AI Index found US corporate AI spending hit a new high in October but grew only 1.65 percent, with one percent of companies driving 80 percent of revenue for OpenAI and Anthropic and customer support emerging as a lower-margin growth driver than coding [27]. By contrast, Aixtron's revenue rose 60 percent quarter on quarter on optoelectronics demand for AI data centres, suggesting hardware supply chains remain buoyant even as software spending concentration and financing stress become visible [25 242725
- Agent security and real-world friction. Microsoft researchers disclosed 'Lazarus', a backdoor technique that reprograms dormant neurons so a normally behaving agent executes malicious commands when a hidden trigger appears, which they describe as especially serious for agents with tool or terminal access [33]. In the same window, United said it will not let personal AI assistants book flights, citing privacy and regulatory concerns, noting it has seen few such attempts so far, in line with DoorDash [11]. Together these point to a gap between the rapid build-out of agent capabilities and the trust, governance and evaluation needed for them to act autonomously in consequential settings [37]. 331137
- AlphaGenome 2 and health-model research. Google DeepMind released AlphaGenome 2, a conditional DNA sequence model predicting regulatory activity and variant effects per cell type, available via the AlphaGenome API for non-commercial research [16][32]. Google Research also described using wearables for cardiovascular-kidney-metabolic risk profiling to support personalised coaching [31]. New arXiv work questioning whether natural-image foundation models reliably beat task-specific medical segmentation models is a useful counterweight to assumptions that general models will simply displace specialised ones [34][35]. 1632313435
Themes
- Enterprise agent plumbing is maturing fast. AWS alone published guidance on running OpenAI Codex via Bedrock, a single-call Web Search API for AgentCore, push-based agent registry updates, reusable generative AI assets and DXC's governed multi-model substrate, while flagging more than 140 AI/ML sessions at re:Invent [3][6][15][28][17][14]. The emphasis on governance, cost reporting and auditability reflects enterprises moving from pilots to operating models, even as Ramp data shows spending growth slowing and highly concentrated [10][27]. 36152817
- Capability sprints outpace trust. Agent-written production software, 1,000-question decision scoring and consumer experiments like Google's Tabletop show how quickly autonomous capability is being deployed [29][20][9]. Yet Microsoft's Lazarus backdoor, United's refusal to accept agent bookings and academic work on agent evaluation stability show trust, security and verification lagging behind [33][11][37][43]. 292093311
- Small-model efficiency research continues. Several arXiv papers examine how to squeeze more from small local models, including escalating weight precision only at uncertain decoding steps and trading precision against option-order averaging under a fixed budget [44][45]. Combined with ReTrace's distillation of tool-using agents into planner-free workflows to cut per-step LLM calls, the research direction is clearly towards cheaper inference at scale, consistent with the price pressure visible in Google's $0.10 per million token decision model [38][12]. 44453812
What to watch
- Whether Nvidia's board formally acts on Huang and how it addresses the national-security dimension, since the dismissal remains a reported intention rather than a confirmed decision [8][2]. 82
- How Jev and OpenAI respond to Google matching Jev's pricing and API design in decision models, which may trigger a price or feature race in high-volume classification [1][12][20]. 11220
- CoreWeave's ability to sustain growth on prepayments and shorter contracts, read against the Ramp index's slowing spending growth and heavy customer concentration for OpenAI and Anthropic [24][27]. 2427
- Whether the AWS Emulator, built without engineer review, proves reliable in practice, and whether vendors respond to the Lazarus backdoor disclosure with new safeguards for agents holding tool or terminal access [29][13][33]. 291333
Stories this analysis is based on (27)
- Gemini Decisions API Simon Willison
- The unfolding Jensen Huang scandal: a timeline TechCrunch
- OpenAI Codex on Amazon Bedrock AWS Machine Learning Blog
- Introducing the Web Search API on Amazon Bedrock AgentCore AWS Machine Learning Blog
- Nvidia is reportedly preparing to fire Jensen Huang The Verge
- Google tests Tabletop, letting groups build planning tools with an AI agent SiliconANGLE AI
- United won’t let your AI agent book your flight The Verge
- Google Launches Gemini Decisions API With Gemini Decision 1 at $0.10/M Tokens MarkTechPost
- aws-emulator Simon Willison
- Automating AWS Agent Registry updates with push-based architecture AWS Machine Learning Blog
- Google DeepMind Releases AlphaGenome 2 for Cell-Type-Specific Variant Effects MarkTechPost
- How DXC Technology built a multi-model AI substrate on Amazon Bedrock AWS Machine Learning Blog
- Google joins the decision-model wave with Gemini Decision 1, scoring 1,000 questions in a single pass The Decoder
- CoreWeave turns to customer prepayments, shorter contracts to fund AI growth TechCrunch
- Aixtron’s revenue grow 60% in Q3, driven by opto demand for AI data centers Semiconductor Today
- US corporate AI spending hits a new high in October, but growth slows to 1.65 percent The Decoder
- Accelerating generative AI delivery with reusable assets AWS Machine Learning Blog
- AWS gives developers a free local stand-in for its cloud, built from scratch by AI agents GeekWire
- Using wearables to enable cardiovascular-kidney-metabolic (CKM) risk profiling for personalized coaching Google Research
- AlphaGenome 2: Deciphering how cells read DNA to better understand disease Google DeepMind blog
- Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents Microsoft Research blog
- Medical Image Segmentation: Foundation Model Adaptation versus Task-Specific Training arXiv (research papers)
- Toward Foundation Detection Models via Example-Based Grounding in EM-DETR arXiv (research papers)
- Comparative Analysis of Agent Evaluation Frameworks: Stability, Detection, and Discovery in Enterprise Analytics Agents arXiv (research papers)
- ReTrace: Distilling Tool-Using Agents into Planner-Free Workflows arXiv (research papers)
- Precision on Demand: Escalating Weight Bits at Uncertain Decoding Steps in a Small Local Model arXiv (research papers)
- Bits or permutations? Spending a fixed inference budget on weight precision or on option-order averaging in small local language models arXiv (research papers)
Written by Claude Fable 5.1 (Anthropic), based on 45 recent headlines and excerpts in this section, Nov 12. AI analysis can misread or miss things: the numbers link to the original reporting, which is the authority.