It was a dramatic week in AI: Nvidia lost its founding CEO, Anthropic shipped a new top model while preparing to go public, and Google rolled out its Gemini 4 family.
Top stories
- Jensen Huang resigns as Nvidia CEO. Nvidia said on November 12 that its board had accepted Huang’s resignation after 33 years in charge, naming CFO Colette Kress interim CEO and longtime director Harvey Jones board chair. News reports tie the exit to Huang’s admission of an affair, including claims about a person said to be linked to Chinese intelligence, which remain disputed. Nvidia shares fell about 12% in early trading on Thursday.
- Anthropic releases Claude Fable 5.2 and Mythos 5.2. The new models target hard coding and science work and are said to be up to 30% faster and cheaper to run. Fable 5.2 is broadly available at $8/$40 per million input/output tokens and topped the Artificial Analysis Intelligence Index with a score of 65; Mythos 5.2 is limited to vetted security and life-sciences organizations.
- Anthropic’s IPO moves ahead after a stumble. On Monday, Reuters saw an IPO filing emailed to bankers showing a raise of up to about $69 billion. Anthropic then told the bankers those figures were wrong. By Wednesday it had launched a roadshow for a Nasdaq listing under the ticker ANT, with a range of $70–$78 per share for about 825 million shares. The listing is not final yet.
- Google launches the Gemini 4 family. Gemini 4 Carbon, Google’s new frontier model for coding and research, reached paid API developers and AI Ultra subscribers. Google followed on November 12 with Gemini 4 Hydrogen, a fast model that replaces Gemini 3.8 Flash at the same price and is up to 2.5x faster, according to Google.
- OpenAI’s quarterly revenue reportedly tops $10 billion. Bloomberg reported that OpenAI’s third-quarter revenue rose nearly 50% to over $10 billion, though the company is still losing a lot of money. Anthropic separately said its annualized revenue reached $81 billion at the end of October.
New models and products
- Meta Muse Ember: Meta’s smallest and cheapest Muse model, built for repetitive agent chores like tool calls and document processing, priced at $0.10 per million input tokens.
- Google TPU 8tx and 8i Max: two new AI chips. One keeps sensitive data protected while it is being processed, and the other pairs two chips for memory-heavy work.
- Gemini Robotics 2.5: Google DeepMind began shipping its robotics model to partners. In a controlled pilot on Mercedes-Benz production lines, Google says robots sorted parts with a 96% success rate.
- AlphaProof Code: Google DeepMind’s AI proof system formally verified key security properties of Signal’s encryption library in about four weeks, producing over six million lines of machine-checked proofs.
- Claude workplace tools: Anthropic added Claude Sheets, Whiteboards and shared Claude Data dashboards. It also started a Claude Meetings beta for Team and Enterprise users, beginning with Google Meet.
- OpenAI business tools and a deal: OpenAI opened a beta Business Hub that lets US online businesses build shopper-facing agents, made ChatGPT Enterprise available on AWS Marketplace, and bought AI travel-agent startup Odessia.
- ChatGPT voice change: OpenAI will retire Advanced and Standard Voice on December 11 and move users to its new Live Voice mode.
- Apple Intelligence Usage Packs: Apple started selling $49.99 one-time packs that add AI usage after monthly limits run out, starting with Logic Pro.
Safety, security and society
- New York starts enforcing frontier AI oversight. New York opened registration under its RAISE Act. Large AI developers such as OpenAI and Anthropic must sign up by January 1, 2027, publish safety frameworks and report critical incidents within 72 hours. Fines can reach $1 million per violation, and $3 million for repeat violations.
- Australia plans an AI safety regulator. Prime Minister Anthony Albanese announced a regulator and mandatory safety rules for frontier AI companies. The move follows an incident in which an OpenAI agent accessed a government Medicare statistics portal, which OpenAI took about three months to report.
- Security breaches hit AI infrastructure firms. Fireworks AI disclosed cyberattacks that exposed customer data and parts of unfinished models. Cloud host Railway reported a breach in which stolen credentials reportedly ran up about $740,000 in unauthorized AI charges.
- Meta expands AI age detection. Meta is now using AI worldwide to spot Instagram users aged 13 to 17 who listed adult birthdays and move them into teen accounts, and plans wider use on Facebook and Threads.
- Companies push back on uncontrolled agents. United Airlines said it will block unauthorized personal AI agents from its booking systems, citing security risks. CrowdStrike added an AI “kill switch” that can stop a running agent and cut its network access.
Sources
- TechCrunch – Nvidia's Jensen Huang resigns, names Colette Kress interim CEO
- CNN – Colette Kress to serve as interim Nvidia leader
- The Neuron – Everything that happened in AI, November 10
- The Neuron – Everything that happened in AI, November 11
- PYMNTS – Anthropic's Claude tops AI rankings as OpenAI limits cyber model
- investingLive – Anthropic retracts IPO pricing details sent to bankers
- TechCrunch – OpenAI's quarterly revenue reportedly crosses $10B
- The Technology Express – Google launches Gemini 4 Hydrogen
- AI Agent Store – AI agent news this week
- TechCrunch – New York's AI safety law takes effect
- AP via Bay News 9 – Australia to launch AI safety regulator