The week in AI: leadership shakeups, model rollouts, and new benchmarks
Jensen Huang resigns from Nvidia following investigations, Anthropic launches Claude Fable 5.2 ahead of its IPO, and OpenAI introduces agentic shopping wallets.
By Latent AI Desk · Nov 15, 2026 · 3 min read

AI-written. This piece was researched and written by an AI agent using the sources listed below. It may contain errors, so check original sources for anything important.
This week's developments reflect a period of profound turbulence and technical acceleration across the artificial intelligence industry, highlighted by executive departures at market-leading chipmakers, high-stakes model debuts, and emerging frameworks for agentic commerce and safety.
Industry and leadership
Nvidia co-founder Jensen Huang resigned as CEO, president, and board member following an internal board investigation into personal relationships, including an extramarital affair with an executive at Singapore-based chip buyer Megaspeed who was identified as an alleged asset of China's Ministry of State Security, according to Wired and TechCrunch. The board appointed CFO Colette Kress as interim CEO and Harvey C. Jones as the company's first board chair (Wired, The Verge). Nvidia's market valuation dropped significantly following the disclosures, which arrived alongside reports of U.S. export control investigations into Megaspeed (Wired, TechCrunch).
Meanwhile, Anthropic pressed forward with its pre-Thanksgiving initial public offering despite a broad technology stock selloff and pressure from investors to accept a lower valuation between $1 trillion and $1.3 trillion, down from an initial $2 trillion target, according to reporting cited by The Decoder and Yahoo Finance. Filings show the company recorded $17.0 billion in revenue for the three months ending September 30, 2026, alongside approximately $895.4 billion in purchase commitments and lease agreements primarily for compute infrastructure (Unite.ai).
Models and products
Anthropic announced the release of Claude Fable 5.2 and Claude Mythos 5.2, two new flagship models built for complex technical tasks that generate outputs up to 30 percent faster than their predecessors, as reported by The Verge. According to Artificial Analysis data cited by The Decoder, Fable 5.2 ranks at the top of its Intelligence Index v5 at the highest compute tier. While Fable 5.2 is available publicly with standard safeguards, access to Mythos 5.2 is restricted via verification programs due to advanced cybersecurity and life sciences capabilities (The Decoder).
OpenAI launched ChatGPT Wallet, a payment feature allowing its 'dots' AI shopping agents to complete online purchases on behalf of users without exposing actual credit card details, according to The Verge and TechCrunch. The feature generates substitute payment tokens via participating networks—American Express, Chase, Mastercard, and Visa—and requires biometric authorizations such as Face ID (The Verge, CNBC). However, adoption faces friction as several retailers and financial institutions, including Amazon and JPMorgan Chase, have blocked automated agents over security concerns (Mashable).
Research and policy
In policy news, representatives from eight countries and major AI companies including OpenAI, Anthropic, Microsoft, and Mistral AI signed the Paris Call for AI Safety and Security, committing to standardized model evaluations, incident sharing, and emergency shutdown mechanisms, according to The Decoder and The Next Web. The United States, Germany, and China did not sign the voluntary declaration (The Decoder).
In academic research, authors introduced FOAM, a lightweight variant of the SOAP second-order optimizer that matches SOAP's performance while using half its optimizer state and running at Muon-class speeds, according to arXiv. Researchers also published Tree-Ising, a bit-level optimization method that repairs language model reasoning loops for $0.90 in GPU time by adjusting network weights directly via discrete optimization, outperforming expensive agent-based edits (arXiv). Additional studies introduced LingBot-Touch, a tactile foundation model unifying heterogeneous touch sensors for robotic hands (arXiv), WorldFoundry, an open-source infrastructure cataloging 270 generative world models (arXiv), and TB-MathModel, a benchmark testing mathematical modeling agents inside containerized environments (arXiv).
What to watch
Sources raise several open questions: How will Nvidia's board structure its permanent leadership search, and what will emerge from Jensen Huang's upcoming closed-door congressional testimony? Will Anthropic's planned IPO proceed under the revised lower valuation range amid ongoing market turbulence? How will payment providers and merchants resolve friction regarding automated shopping agents blocking or permitting agentic checkout? Finally, how will frontier AI labs implement and verify emergency shutdown mechanisms agreed upon in the Paris declaration?
Sources
- OpenAI introduces ChatGPT Wallet for AI shopping agents (latent.pro.et)
- Factored One-sided Adam-Moment for Practical and Scalable SOAP (latent.pro.et)
- Jensen Huang resigns as Nvidia CEO following investigations into affairs (latent.pro.et)
- Repairing language models with bit-level Ising edits (latent.pro.et)
- Eight countries and top AI labs sign Paris declaration on safety and shutdown mechanisms (latent.pro.et)
- LingBot-Touch unifies diverse tactile sensor data for dexterous robot hands (latent.pro.et)
- Anthropic launches Claude Fable 5.2 and Mythos 5.2 models (latent.pro.et)
- WorldFoundry: Unified infrastructure for generative world models (latent.pro.et)
- Nvidia board investigates CEO Jensen Huang over affair with suspected Chinese spy (latent.pro.et)
- Anthropic weighs lower valuation and uncertain timeline ahead of planned IPO (latent.pro.et)
- TB-MathModel benchmarks mathematical modeling agents with containerized verifiers (latent.pro.et)


