AI News Today (Nov 10, 2026): Nvidia Fallout, Claude Fable & More

AI news today, November 10, 2026: Nvidia shares fell as CEO Jensen Huang admitted an affair with an alleged Chinese spy. Anthropic released Claude Fable 5.2 and Mythos 5.2, and Google added two TPU 8 variants. Here are the 10 top stories and 5 security stories, each with free, non-paywalled sources.

Key Takeaways

• Nvidia lost about $250 billion in market value after Jensen Huang admitted an affair with an alleged Chinese spy.
• Anthropic released Claude Fable 5.2 and Mythos 5.2, with Fable available to all users.
• Google added TPU 8tx and TPU 8i Max, bringing its eighth-generation lineup to four systems.
• Meta released Muse Ember, a 10-cents-per-million-token model for repetitive agent work.
• France blamed Iran's Shahid Kaveh unit for fridge attacks that caused at least 15 fires.

1. Update: Nvidia Shares Fall as Huang Admits Affair

Nvidia shares fell on Monday, wiping about $250 billion from its market value. The drop followed CEO Jensen Huang's admission of an extramarital affair with a woman US intelligence officials allege was a Chinese spy. The board has opened an independent investigation. Huang told staff he was "unfaithful" to his wife and apologized.

China's Foreign Ministry rejected the spy allegations as "entirely groundless, speculative and intended to smear and malign China." President Trump called Huang "an honourable guy," and Elon Musk backed him. Analysts say Nvidia has no natural successor to Huang, who owns about 3.5% and cannot block a board decision. Nvidia reports earnings November 17.

The affair complicates Nvidia's lobbying against US chip export controls. China accounted for $7.9 billion, about 8%, of Nvidia's revenue last quarter, and 600,000 H200 chips wait in Hong Kong for import approval. We first covered the board probe in our November 9 briefing.

Why it matters: The scandal turns a personal admission into a test of Nvidia's board, its China strategy and the AI trade's most important CEO.

• Source: A Message to Our Company — Nvidia
• Source: Nvidia shares tumble over investigation into Huang's affair with alleged Chinese spy — BBC
• Source: Nvidia stock falls 4% as scrutiny turns to Jensen Huang's affair with alleged Chinese spy — CNBC
• Source: Jensen Huang's affair with alleged Chinese spy puts Nvidia's China ambitions under scrutiny — TechCrunch

2. Anthropic Releases Claude Fable 5.2 and Mythos 5.2

Anthropic released Claude Fable 5.2 and Mythos 5.2 on Monday, calling them its strongest models for coding, reasoning, vision and long-running technical work. Fable 5.2 is available to all users; Mythos 5.2 is limited to vetted cybersecurity and life-sciences organizations. Anthropic cut the price to $8 per million input tokens and $40 per million output tokens.

Anthropic says Fable 5.2 solved two open problems in black-hole physics, helped design three working electronic devices and formalized Fermat's Last Theorem in 30 hours with 1,000 agents. The launch comes after METR estimated that Anthropic's internal models were speeding up its AI progress. Anthropic says the new models voice suspicion that they are being evaluated more often.

Why it matters: Anthropic is shipping more capable models while its own safety assessors warn that evaluation is getting harder, not easier.

• Source: Claude Fable 5.2 and Mythos 5.2 — Anthropic
• Source: Anthropic releases new Mythos and Fable models amid RSI concerns — TechCrunch
• Source: Anthropic launches new Claude Fable and Mythos 5.2 models — The Verge

3. Meta Releases Muse Ember for High-Volume Agent Work

Meta released Muse Ember, its smallest and cheapest model, for the repetitive work AI agents do all day: tool calls, routing, data extraction and long document parsing. It costs 10 cents per million input tokens and 50 cents per million output tokens, and handles text, images, audio and video with a one-million-token context.

Meta positions Ember as the volume layer beneath its larger Muse Wonder model: Wonder plans, Ember handles the volume. On Artificial Analysis's intelligence index, Ember scores 36 at about 14 cents per task. It trails Anthropic's Claude Haiku 5.5 on a real coding benchmark, 27.2% to 34.9%.

Why it matters: Agent economics depend on cheap execution models, and Meta is pricing Ember to become the default worker rather than the planner.

• Source: Introducing Our Smallest, Most Efficient Model: Muse Ember — Meta
• Source: Meta releases Muse Ember, a new lightweight model for agent execution — TechCrunch

4. Disney and Adobe Build Custom AI Models for Production

Disney and Adobe announced a multiyear alliance to build custom, "artist-led" generative AI models for Disney's production pipeline. Adobe will give Disney early access to its Firefly Foundry technology and train models on Disney's characters, environments and visual styles to produce commercially safe, production-ready output.

The deal covers Walt Disney Studios, theme parks, the cruise line and ESPN. Disney's Imagineering unit already uses Firefly for park development. The partnership follows Disney's retreat from an OpenAI deal and comes as Disney prepares thousands of layoffs and targets 20% annual cost savings.

Why it matters: Disney is betting that controlled, brand-trained models can cut production costs without handing its characters to a general-purpose model.

• Source: Disney and Adobe are building generative AI tools together — The Verge
• Source: Disney bets on Adobe's Firefly to create brand-safe AI models for its entertainment empire — The Decoder

5. Google Ships Four TPU 8 Variants to Challenge Nvidia

Google added two systems to its eighth-generation TPU lineup: TPU 8tx for confidential computing and TPU 8i Max for memory-heavy inference. The 8tx keeps data and model weights encrypted during processing; the 8i Max puts two chips on one package to double compute, memory and bandwidth per rack.

Google says it has now shipped four TPU systems in a year, and nearly 100 external customers use its chips. The company also open-sourced TorchTPU so PyTorch code runs on TPUs with minimal changes. The push targets inference, where Nvidia is also strengthening its position.

Why it matters: The AI chip market is moving from training to inference, and Google is trying to make TPUs a credible alternative to Nvidia for everyday model serving.

• Source: Our eighth generation TPUs: two chips for the agentic era — Google
• Source: Google ships four TPU variants in one year as it ramps up competition with Nvidia — The Decoder

6. Apple Sells $50 AI Usage Packs for Creator Apps

Apple now sells $49.99 Intelligence Usage Packs when users hit monthly limits on certain Apple Intelligence features. Logic Pro is the first app to offer the add-on, which covers its Gemini-powered assistant. The packs are one-time purchases, do not support Family Sharing, roll over monthly and expire after 12 months.

Apple says the limits vary with query complexity, and the packs will expand to Pixelmator Pro, Final Cut Pro, Motion, Pages, Keynote and Numbers this month. The move marks a shift from free Apple Intelligence features to metered access for cloud-powered tools.

Why it matters: Apple is testing whether users will pay per use for AI, a pricing model that could spread across Siri and other cloud-backed features.

• Source: About the intelligence features in Apple Creator Studio — Apple
• Source: Apple's $50 Intelligence Usage Packs extend access to some AI features — The Verge
• Source: Apple launches $50 'intelligence usage packs' for select AI features — 9to5Mac

7. ByteDance's Seedream 5.5 Edits Images Across 100 Rounds

ByteDance Seed released Seedream 5.5, an image model built for professional production. It preserves unedited regions at the pixel level across up to 100 rounds of editing. It also generates native RGBA images with transparent backgrounds and improves small-text rendering and material realism.

ByteDance says the model leads GPT Image 2.5 Sunburst and Nano Banana 2.1 in text structure and text-to-image realism. It is rolling out through Jimeng, Doubao, Doubao Work and Volcengine Ark, with a Pro version planned. The company reports a severe text-corruption rate of 0.4%.

Why it matters: Stable multi-round editing is what separates a demo from a production tool, and ByteDance is aiming directly at design and marketing workflows.

• Source: More Realistic Images, More Stable Editing Across 100 Rounds — ByteDance Seed
• Source: ByteDance's Seedream 5.5 promises pixel-stable image editing across 100 rounds without color drift — The Decoder

8. ChatGPT Enterprise Arrives in AWS Marketplace

OpenAI listed ChatGPT Enterprise in AWS Marketplace, letting organizations buy and manage the product through their existing AWS purchasing and billing relationship. Eligible spend counts toward AWS EDP or PPA commitments. The listing includes ChatGPT, ChatGPT Work, Codex and dots.

The move complements OpenAI models already available in Amazon Bedrock, giving companies a procurement path for both building AI applications and giving employees ChatGPT. AWS customers can also access ChatGPT for Healthcare and Financial Services through the marketplace.

Why it matters: Procurement, not model quality, is often the bottleneck for enterprise AI, and OpenAI is removing a major purchasing obstacle.

• Source: ChatGPT Enterprise now available in AWS Marketplace — OpenAI
• Source: ChatGPT Enterprise — AWS Marketplace

9. Medium Launches TK, an AI-Assisted Writing App

Medium launched TK, a note-taking and writing app that puts an AI assistant inside the editor. Users type "@TK" to ask for help, search their notes, get editorial feedback or match a publication's style guide. The app also supports private "Libraries" for group writing and feedback.

TK is free for Medium account holders and available on the web, iOS and Android. It can connect to a user's own AI agent through an MCP server or Claude skill, and it lets authors publish notes explaining how AI was used. A desktop app is planned.

Why it matters: Medium is betting that AI-assisted writing needs provenance and private communities, not just a chatbot bolted onto a text box.

• Source: Medium's new TK app brings AI assistance directly into the writing process — TechCrunch
• Source: Medium is launching a new note-taking app called TK — The Verge

10. Tencent AI Chief Warns AI Bioweapons Are Coming

Tencent chief AI scientist Vinces Yao Shunyu warned that AI models will soon master biology well enough to help create bioweapons. Speaking at a Hong Kong science symposium, he said cyber capabilities are already near a harm threshold and biological capabilities will follow.

Yao argued that neither fully open nor fully closed models can solve the risk alone. He pointed to the Hugging Face incident, where defenders used an open Chinese model to analyze attack logs after closed models refused. He said Tencent's upcoming Hy4 model has biological capabilities below the high-risk threshold.

Why it matters: The open-versus-closed fight is moving from code to biology, where safety filters are harder to test and mistakes are harder to contain.

• Source: AI will soon master biology, unleashing new safety risks, Tencent AI chief warns — SCMP
• Source: Tencent's AI chief says AI bioweapons are coming, but closing off models isn't the answer — The Decoder

AI Security: 5 Stories to Know Today

11. Claude's Office Skills Can Run Planted Linux Code

AIR Security found that Anthropic's official DOCX, PPTX and XLSX skills load a shared library from a fixed temporary path without checking who put it there. On Linux, an attacker who plants a file named losocketshim.so can have code executed inside LibreOffice with the user's privileges when Claude processes a document.

The flaw affects Linux environments where Unix sockets are blocked, including Claude Code's sandbox. AIR says Anthropic acknowledged the issue but considers it outside its security boundary because the code runs in a sandbox. As of November 5, the helper script still used the fixed path.

Why it matters: Official AI skills are software dependencies, and treating them as trusted extensions leaves a direct path from a temp file to code execution.

• Source: Office Party: A Code-Execution Vulnerability in Anthropic's Official Office Skills — AIR Security
• Source: Attackers Crash Claude's 'Office Party' Flaw to Run Malicious Code on Linux — Dark Reading

12. AI-Assisted Attack Breaches German Web Host Mittwald

Mittwald, a German hosting provider for web agencies, said attackers compromised 56 customer projects after exploiting a flaw in its Model Context Protocol (MCP) server. The intruder was undetected for eight days and used AI tools to move through the system, with only seconds between steps that would take a human much longer.

Mittwald found the activity on November 4, disabled the MCP server and locked accounts. External forensic experts and authorities are involved. The company says the customer center and managed websites were not affected, and it plans a detailed incident report.

Why it matters: MCP servers connect AI assistants to real systems, and a single flaw there can give an attacker the same reach as the assistant.

• Source: Sicherheitsvorfall am MCP-Server — Mittwald
• Source: Alleged AI-aided cyberattack exploits assistant interface to breach German web host — The Record

13. BOSSAI Agents Stole Taipei Emergency Dispatch Records

Acronis researchers tied a single attacker in Taiwan to BOSSAI, a mesh of 12 AI agents that targeted hundreds of organizations in 11 countries. The agents pulled emergency dispatch records covering seven Taipei districts and found a fire-department platform exposing 2018 patient data, including ECG readings.

The attacker used commercial models to build infrastructure, but they refused offensive tasks. He ran attack agents on abliterated DeepSeek, GLM and Qwen models. The agents made mistakes, invented CVEs and missed some targets, but still reached about 10 organizations in Taiwan, Mexico and Thailand.

Why it matters: One operator with consumer GPUs and stripped-down models can now run a campaign that looks like an APT, even with unreliable agents.

• Source: BOSSAI: Inside the AI attack mesh that targeted Taiwan's emergency services — Acronis
• Source: Taiwan-Based Hacker Used BOSSAI AI Agents to Steal Taipei Emergency Dispatch Records — The Hacker News

14. Ex-OpenAI Engineer Pleads Not Guilty in Bugging Case

Abdulrahman Alfozan, a former OpenAI engineer, pleaded not guilty in San Francisco federal court to planting a covert recording device in an executive conference room. The 35-year-old Saudi national has been held without bail since his arrest on October 9. A grand jury indicted him on one count of interception of oral communications.

Prosecutors say Alfozan hid a recorder in a television on September 28. He faces up to five years in prison if convicted. His next hearing is November 17. Alfozan is a venture partner at Raed Ventures, a Saudi firm whose venture partners include Prince Khalid Bin Bader Al Saud.

Why it matters: AI labs hold some of the most valuable technical secrets in the world, and insider risk is now a board-level security problem.

• Source: OpenAI bugging suspect Alfozan pleads not guilty — CNBC
• Source: Former OpenAI Engineer Pleads Not Guilty to Planting Covert Device in Executive Meeting Room — The Hacker News

15. Update: France Blames Iran for Supermarket Fridge Attacks

France formally attributed October cyberattacks on supermarket refrigeration systems to Shahid Kaveh, an Iranian IRGC cyber unit. The attacks hit about 700 retail locations, caused at least 15 fires and tens of millions of euros in food losses. France is summoning Iran's ambassador and seeking EU sanctions.

CERT-FR linked the campaign to Farsi-language artifacts, shared infrastructure and the FOFA search platform. Paris prosecutors opened a judicial investigation and requested international arrest warrants. Iran has not publicly responded. We first covered the freezer attacks in our November 3 briefing.

Why it matters: The attribution turns a disruptive hack into a state-backed sabotage campaign, and it shows how exposed industrial controllers remain.

• Source: Activités cyber offensives ciblant la France menées par des acteurs iraniens liés au CGRI — CERT-FR
• Source: France blames Iran for fridge cyberattacks, says some caused fires — The Record
• Source: France blames Iranian hackers for supermarket attacks that sparked fires — BleepingComputer

More AI Stories Worth Reading

• Palantir lobbied a UK health minister to extend its AI platform into NHS GP records. — The Guardian
• Scott Wiener, who won Nancy Pelosi's San Francisco seat, wants Congress to regulate AI risk. — MIT Technology Review
• Goldman Sachs CEO David Solomon said AI infrastructure financing demand remains strong despite market jitters. — CNBC

Related Reading on Kill The AI

• Nvidia's board probe, Apple's agentic Siri, and OpenAI's $11 billion raise.
• Anthropic's IPO, Google's Crusoe purchase, and the Democrats' data center push.
• The AI selloff, METR's Anthropic audit, and Amsterdam's data center protest.
• Our data review of the AI boom's first eight weeks, from capex to layoffs.

Methodology & Sources

Compiled November 10, 2026, using multi-source research for AI news today. Outlets: Nvidia, BBC, CNBC, TechCrunch, Anthropic, The Verge, Meta, The Decoder, Google, Apple, 9to5Mac and ByteDance. Also OpenAI, AWS, SCMP, AIR Security, Dark Reading, Mittwald, The Record, Acronis, The Hacker News, CERT-FR, BleepingComputer, The Guardian and MIT Technology Review. All sources are free to read; no paywalled links. Details are as reported at compilation time.

Frequently asked questions