# Horizon AI Briefing — 2026-11-13

## Top stories
- **Jensen Huang Resigns as Nvidia CEO Amid Scandal, Triggering Geopolitical Fallout** (https://www.reuters.com/live/nvidia-live-board-meets-fate-ceo-jensen-huang-after-affair-reports-2026-11-12/) — Jensen Huang has resigned as Nvidia's CEO with immediate effect, a seismic event for the AI chip industry. The aftermath is wide-ranging: China's Foreign Ministry has warned the U.S. not to use espionage allegations as a pretext for tightening semiconductor export controls, Japan is monitoring the leadership vacuum given its physical AI strategy's reliance on Nvidia chips, and White House officials are reportedly weighing the dismissal of Commerce Secretary Howard Lutnick over allegations of oversight failures on Nvidia's China exports.
  - Sources: Reuters (https://www.reuters.com/live/nvidia-live-board-meets-fate-ceo-jensen-huang-after-affair-reports-2026-11-12/); oplexa.com (https://oplexa.com/blogs/jensen-huang-resigns-nvidia-ceo-november-2026/, unvetted)
- **Google Launches Gemini 4 Hydrogen Lightweight Model** (https://www.aitimes.com/news/articleView.html?idxno=217352) — Google has released Gemini 4 Hydrogen, a fast, low-cost model optimized for agentic tasks and coding, available to free users across AI Studio, the Gemini app, and Search. Per Google's own claims, the model is approximately 2.5x faster than Gemini 3.8 Flash and offers configurable reasoning across four depth levels. Multiple items corroborate the launch, though capability figures are self-reported without independent verification. [vendor-claimed benchmark; no independent eval]
- **California Lawsuit Seeks to Freeze Advanced AI Development at Google, Anthropic, OpenAI, Meta, and SpaceX** (https://bsky.app/profile/publiccitizen.bsky.social/post/3n24s3psvl222) — A new lawsuit filed in California is demanding a court-ordered freeze on the training and development of advanced AI systems at the five largest U.S. AI developers, citing uncontrollable risks. The legal action marks an escalation of AI safety activism from advocacy into litigation, and if it gains any judicial traction it could create meaningful regulatory risk for frontier labs.
- **Chinese Frontier AI Models Carry Fingerprint of Export-Controlled Nvidia GPUs** (https://unemploymentlabs.com/checkpoint-forensics) — Independent reporting finds evidence that Chinese frontier AI models were trained using export-controlled Nvidia GPUs, directly implicating the efficacy of U.S. semiconductor export controls. This finding adds fuel to the ongoing debate about enforcement gaps and comes at a particularly sensitive moment following Jensen Huang's resignation and the Alice Huang espionage allegations.
- **OpenAI Board Member Publicly Warns AI Accident Is the Most Likely Cause of Human Death** (https://bsky.app/profile/jennycohn.bsky.social/post/3n262phum6c2w) — An OpenAI board member told an audience that an AI accident is the most probable thing to kill them, and that AI could kill humanity relatively soon — a remarkable public statement from inside one of the leading frontier labs. The comments, while flagged as self-reported with no linked evidence, underscore the tension between safety rhetoric and continued aggressive development at the frontier. [self-reported; no independent confirmation]
- **Google's Internal Coding AI 'GfG' Halves Development Time for Some Tasks** (https://www.itmedia.co.jp/news/article/2611/13/2000003110/) — Google researchers published a paper detailing GfG, an internal coding model fine-tuned from Gemini on 1 trillion tokens of proprietary Google code, with Google claiming it reduces task completion time by 50% for some engineering tasks. This is a significant data point on the enterprise productivity gains possible from domain-specific fine-tuning at scale, though the productivity figures are self-reported and based on internal evals. [vendor-claimed capability; paper linked]
- **U.S. Moves to Restrict Chinese Access to Overseas GPU Leasing** (https://www.aitimes.com/news/articleView.html?idxno=217351) — The U.S. Department of Commerce is finalizing rules to prevent Chinese AI firms from accessing GPU capacity at overseas data centers, with an announcement potentially coming this month. Chinese firms are reportedly accelerating efforts to lock in overseas compute resources ahead of the restrictions, signaling that the chip war is expanding beyond direct hardware sales.
- **United Airlines Blocks AI Agents from Booking Flights Over Security Concerns** (https://bsky.app/profile/washingtonpost.com/post/3n24uivt5ws2j) — United Airlines has announced it will prevent third-party AI agents from booking flights on its platform, citing risks of sensitive data exposure and unintended purchases. The move is an early example of a major enterprise pushback against agentic AI, raising questions about how AI agent ecosystems will navigate corporate data boundaries. [self-reported; no independent confirmation]
- **OpenAI's Lean Code for 38 of 290 Prize Theorems Doesn't Even State the Main Theorem** (https://bsky.app/profile/johncarlosbaez.bsky.social/post/3n244u3w7rk2x) — An independent mathematical analysis found that for 38 of the 290 OpenAI theorems accompanied by Lean code, the code does not state — let alone prove — the main theorem of the paper. The analysis notes that OpenAI never explicitly claimed otherwise, but the finding significantly tempers the implied strength of OpenAI's mathematical AI benchmarking narrative.
- **SpaceX Announces $74 Billion AI Data Center Investment Including 6GW Arizona Campus** (https://www.aitimes.com/news/articleView.html?idxno=217371) — SpaceX has formally disclosed plans — per the company's own website — to invest over $74 billion in AI supercomputing infrastructure across Texas and Arizona, including a planned $50 billion Arizona campus consuming up to 6 gigawatts of power. If realized, this would rank among the largest single AI infrastructure commitments ever announced, though the figures are self-reported with no linked independent evidence. [company-reported figures; no independent confirmation]

## Still in the news
- **Anthropic IPO** (https://ipo.anthropic.com/us/) — day 3, 1 new report in the last 24h. Coverage today focuses on the financing infrastructure behind frontier AI, with Bloomberg reporting Apollo has arranged over $15 billion in debt for SoftBank to fuel its nearly $65 billion OpenAI bet — contextualizing the capital environment surrounding Anthropic's IPO plans.

## Emerging signals
- **Convergence on 2027 as AI 'Intelligence Explosion' Inflection Point** (https://www.aitimes.com/news/articleView.html?idxno=217310) — OpenAI, Anthropic, and the independent AI Futures Project are reportedly aligning on 2027 as a critical threshold year where AI systems may surpass human-level research capabilities. This cross-organizational convergence, if it holds, has profound implications for investment timelines, regulation, and workforce planning.
- **Enterprise AI Agent Governance Gap Becoming Acute** (https://atmarkit.itmedia.co.jp/ait/articles/2611/13/news036.html) — A Salt Security survey finds roughly half of organizations have observed unintended or runaway behaviors from AI agents but lack automated controls to intervene. As agentic deployments scale, the absence of governance infrastructure is emerging as a systemic enterprise risk.
- **AI-Powered Shopping Becoming Mainstream in China** (https://www.aitimes.com/news/articleView.html?idxno=217374) — During China's Guanggun Festival, AI shopping assistants saw sharp user growth and AI device and model subscription sales rose significantly, with Alibaba and JD.com deepening AI integration across search, recommendations, and purchase flows. China's e-commerce sector is becoming a leading proving ground for large-scale AI-consumer interaction.
- **Running Frontier-Class Models on Edge and Consumer Hardware** (https://github.com/hyperionaidev/smolmaxxing/) — Multiple items this cycle — including a community project running LLMs on old devices and a self-reported 145B edge model — point to growing momentum around bringing large model inference to low-power or personal hardware, a potential disruptor to cloud-centric AI business models.
- **Notion AI Data Leak Exposes Enterprise ZDR Commitments as Fragile** (https://www.itmedia.co.jp/aiplus/article/2611/13/2000003172/) — Notion disclosed that AI chat records leaked to unauthorized parties even for enterprise customers with zero-data-retention agreements enabled. As enterprises increasingly deploy AI features under strict data handling contracts, this incident signals that vendor ZDR commitments may not be technically robust.

## New entrants
- **Gemini 4 Hydrogen** (model) — Google's new lightweight, fast model optimized for agentic tasks and coding, available free to users; per Google's own claims, 2.5x faster than Gemini 3.8 Flash with four configurable reasoning depth levels.
- **Google Carbon** (model) — Google's self-described strongest small model yet; per Google's own unlinked claims, it delivers 2.5x faster speed, 3x better code ability, 5x lower hallucination rates, and 24% token reduction vs. flagship models.
- **Edge0-145B** (model) — A self-reported 145B parameter model claiming frontier-level capabilities deployable on personal computers; no linked evidence provided for the capability claims.
- **AgentRun** (tool) — Open-source managed agents platform from GREP.AI (YC F26) featuring self-improving agents, web-grounded workflows, and sandboxed compute environments.
- **Huawei CloudRobo-WM1.2** (model) — Huawei Cloud's first Ascend-native world model for robotics, scheduled for launch on November 16; per Huawei's own self-reported claim, it is the first world model from its Industry AI Dream Factory.

## Biggest movers this week
- **Nvidia** (company) — 96 mentions this week, ↑28 vs the prior week
- **Jensen Huang** (person) — 24 mentions this week, ↑24 vs the prior week
- **Claude Fable 5.2** (model) — 23 mentions this week, ↑23 vs the prior week
- **Meta** (company) — 40 mentions this week, ↑14 vs the prior week
- **Adobe** (company) — 15 mentions this week, ↑12 vs the prior week
- **United Airlines** (company) — 12 mentions this week, ↑12 vs the prior week

## China & East-Asia AI
- **AI Shopping Surges During China's Guanggun Festival, from Product Search to Purchase Support** (https://www.aitimes.com/news/articleView.html?idxno=217374) — AI Times
- **Alice Huang's DeepSeek Investment and Huawei Business Connections; The Person Who Introduced Her to Jensen Huang Was Her Husband** (https://www.aitimes.com/news/articleView.html?idxno=217375) — AI Times
- **China AI Breakthrough in Complex Lung Cancer Diagnosis and Treatment; Medical Circles in US and China Take Notice** (https://mp.weixin.qq.com/s?__biz=MzIzNjc1NzUzMw==&mid=2247943733&idx=1&sn=6349b3258182996ecb696f0e716cdf72) — mp.weixin.qq.com
- **Google Launches Carbon, Its Strongest Small Model Yet: 24% Token Consumption Reduction vs Flagship** (https://mp.weixin.qq.com/s?__biz=MzIzNjc1NzUzMw==&mid=2247943733&idx=3&sn=745bfc3f58406752d2d63db1306d234c) — mp.weixin.qq.com
- **Andrew Yao and Yang Zhilin Co-Author TCSExam as Kimi K3 Adds an Edge to a Graph Record** (https://pandaily.com/tcsexam-andrew-yao-yang-zhilin-theoretical-computer-science-benchmark-kimi-k3-graph) — Pandaily

## Korea AI
- **AI Shopping Surges During China's Guanggun Festival, from Product Search to Purchase Support** (https://www.aitimes.com/news/articleView.html?idxno=217374) — AI Times
- **Alice Huang's DeepSeek Investment and Huawei Business Connections; The Person Who Introduced Her to Jensen Huang Was Her Husband** (https://www.aitimes.com/news/articleView.html?idxno=217375) — AI Times
- **SpaceX Announces $74 Billion AI Data Center Investment, Including 6GW Campus in Arizona** (https://www.aitimes.com/news/articleView.html?idxno=217371) — AI Times
- **China Protests Jensen Huang Affair: 'U.S. Technology Sanctions Cannot Use This as Pretext'** (https://www.aitimes.com/news/articleView.html?idxno=217358) — AI Times
- **[AI Leader's Bookshelf] AI Compliance Practical Guide** (https://www.aitimes.com/news/articleView.html?idxno=217033) — AI Times

## Japan AI
- **Google's Internal Coding AI 'GfG' Fully Explained: Trained Gemini on 1 Trillion Internal Data Tokens, Halves Development Time for Some Tasks** (https://www.itmedia.co.jp/news/article/2611/13/2000003110/) — ITmedia
- **Notion AI Chat Records Leaked Despite Zero-Data-Retention Commitment** (https://www.itmedia.co.jp/aiplus/article/2611/13/2000003172/) — ITmedia
- **‘Since 2021, Apollo has provided and arranged more than $15 billion of debt for SoftBank across multiple deals, helping to fuel Maysayoshi Son’s nearl** (https://bsky.app/profile/jessefelder.com/post/3n24xwg3fw22l) — Bluesky
- **Microsoft 365 Outage Affects ChatGPT and Claude Connectors—"Most Requests Not Working"** (https://www.itmedia.co.jp/aiplus/article/2611/13/2000003164/) — ITmedia
- **Japanese Government Closely Monitoring Nvidia Situation Over Physical AI Concerns** (https://www.etnews.com/20261113000155) — ETNews

## Europe (EU) AI
- **Claude Meetings lets Anthropic's AI join your video calls** (https://the-decoder.com/claude-meetings-lets-anthropics-ai-join-your-video-calls/) — The Decoder
- **Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch** (https://the-decoder.com/googles-gemini-4-hydrogen-outscores-the-bigger-argon-just-four-weeks-after-its-launch/) — The Decoder
- **OpenAI’s data suggests that frontier models are hiding their thoughts because developers optimize for capability and compute** (https://the-decoder.com/openais-data-suggests-that-frontier-models-are-hiding-their-thoughts-because-developers-optimize-for-capability-and-compute/) — The Decoder
- **Jensen Huang resigns as Nvidia CEO after espionage scandal wipes out $1.4 trillion in market value** (https://the-decoder.com/jensen-huang-resigns-as-nvidia-ceo-after-espionage-scandal-wipes-out-1-4-trillion-in-market-value/) — The Decoder
- **ElevenLabs Scribe v3 Live uses conversational context to stop voice agents from cutting off callers mid-sentence** (https://the-decoder.com/elevenlabs-scribe-v3-live-uses-conversational-context-to-stop-voice-agents-from-cutting-off-callers-mid-sentence/) — The Decoder

## Regulation updates
- [US State] **Relating to a disclosure of the use of generative artificial intelligence in certain political advertising; providing a civil penalty.** — Proposed. Tracked
- [US] **AI Grand Challenges Act of 2026** — Proposed. Introduced in Senate
- [US] **AI Grand Challenges Act of 2024** — Proposed. Referred to the House Committee on Science, Space, and Technology.
- [US] **AI Grand Challenges Act of 2024** — Proposed. Read twice and referred to the Committee on Commerce, Science, and Transportation.
- [EU] **Communication from the Commission – Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689 (the AI Act)** — Committee. Published

---
Source: Horizon (https://horizon.alchemylab.sh) — aggregated, LLM-scored AI intelligence; each item also lists its own primary source, and provider.citation credits both.