News of the week: November 2–8, 2026 (week 45)
The week was dominated by the US elections and their impact on the AI industry: after the Democrats won the midterms, calls emerged to pause data center construction and AI stocks fell sharply. OpenAI responded with a pledge to local communities, closed the first part of a funding round at a $1.25 trillion valuation and introduced the GPT-6 Cyber security model. Companies raced to build tools for programmers, with OpenAI buying Pierre and Meta acquiring the startup Conductor. Anthropic updated its IPO filing, China is speeding up work on a comprehensive AI law, and the week closed with a scandal involving Nvidia CEO Jensen Huang.
New models
OpenAI introduces GPT-6 Cyber for finding vulnerabilities
On November 5, OpenAI introduced GPT-6 Cyber, a specialized model derived from GPT-6 Astra. It is designed to find vulnerabilities in software, verify whether they can be exploited and propose fixes, but only within the scope that a security team authorizes.
According to the company, researchers using it found 81 previously unknown serious bugs in Firefox and created working exploits (programs that take advantage of a bug) for 948 vulnerabilities in the Linux kernel. Under its Preparedness Framework, OpenAI rates the model as “critical”, so for now it is available only to selected verified customers through the API and Codex (OpenAI).
Google releases Gemini Omni 1.2 Flash for video creation
On November 5, Google made available Gemini Omni 1.2 Flash, a model that generates and edits video. Maximum clip length has grown from 10 to 15 seconds, and videos can be edited conversationally, through step-by-step instructions in natural language.
The model is available through the Gemini API, in the Gemini for business offering and in Google Vids. Outputs carry an invisible SynthID watermark that shows the video was created by artificial intelligence (Google Blog).
Images 2.6: ChatGPT can make animated GIFs
On November 4, OpenAI released Images 2.6, a new version of the image generator in ChatGPT and Codex. The headline feature is the creation of animated GIFs from a text prompt or an image, including control over speed and individual frames.
- According to the company, generation is up to 50% faster.
- Developers get two new models in the API: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst.
- According to OpenAI, more than 4 billion images are created every week through ChatGPT and the API (Digit).
Products and agents
Claude analyzes personal finances
On November 5, Anthropic launched a beta of its Personal Finance feature. After users link their bank and card accounts through Plaid, Claude answers questions about their spending.
Access is read-only, so Claude cannot send or invest money. For now the feature is available only to paying subscribers in the US (TechCrunch).
Apple tests Apple Intelligence in China with the Qwen model
Since November 4, Apple has been testing Apple Intelligence features in mainland China in a developer beta of iOS. Some features, such as writing tools and visual recognition, are powered by Alibaba‘s Qwen model in order to comply with Chinese rules.
A broader launch is planned for December with the iOS 27.3 update (Euronews, TechNode).
Google turns off AI in school accounts
On November 3, Google announced that Gemini, Gemini in Classroom and also Google Search will be turned off by default for student accounts managed by schools. A school administrator has to turn them on.
The catch is that a school that disables AI completely cannot offer students search either, because it is inseparably tied to AI summaries. With this, Google reverses its August approach, when it turned on Gemini in Classroom automatically (TechCrunch).
- November 2: OpenAI described Pando, its internal system for storing code in git, which cut median push latency by roughly 80% after a sharp rise in commits from Codex (AI Weekly).
- November 5: The startup Aina unveiled a $399 smart ring with a display, a joystick and a microphone for controlling AI agents; shipments are due to start in January 2027 (TechCrunch).
Business and investment
AI stock sell-off after the elections
In the US elections on November 3, the Democrats gained the upper hand, and right afterward their leaders called for limits on building data centers. Senator Bernie Sanders wrote on X “National data center moratorium now”, calling for a nationwide halt to construction (CNBC).
Markets reacted sharply: the S&P 500 index fell 4.4% on November 5, its biggest drop since April 2025, and shares of Nvidia and other companies tied to AI infrastructure weakened significantly (The Next Web, CNBC).
OpenAI: a pledge to communities and the first part of a new round
On November 5, OpenAI published its Community Compact. In it, the company promises to cover all electricity and grid connection costs for the data centers it owns, so that neighbors’ bills do not rise, and to stop signing non-disclosure agreements (NDAs) with officials (Android Headlines).
According to Bloomberg, by November 6 the company had raised $11 billion as the first part of a funding round at a $1.25 trillion valuation. The investors reportedly include SoftBank and the Emirati fund MGX; OpenAI has not officially confirmed the terms (GV Wire / Bloomberg).
The battle for programmers: OpenAI buys Pierre, Meta buys Conductor
On November 4, OpenAI announced the acquisition of the small company Pierre Computer, maker of Code Storage, a service for storing code in git and libraries that already runs inside the Codex app. The acquisition is still awaiting regulatory approval (OpenAI).
The same day, Meta bought the startup Conductor, whose app lets several coding agents work at once, each in a separate copy of the code. The Conductor Cloud service will shut down on November 20; the price was not disclosed (Axios, SiliconANGLE).
Anthropic updates its IPO filing
On November 2, Anthropic filed an amended S-1/A prospectus for its planned listing on the Nasdaq. It added results through September 30, 2026, which show quarterly revenue rising to $17 billion.
Among its risk factors, the company now states that its next models may soon approach the threshold it uses itself for “dramatically accelerated” AI progress, meaning a state in which AI significantly speeds up its own development (Crypto Briefing, TechBuzz).
- November 5: Cat Wu, who led the Claude Code product at Anthropic, moved to OpenAI, where she will lead product management for the Core Products group (WinBuzzer).
Regulation and policy
China speeds up a comprehensive AI law
On November 3, China’s National Development and Reform Commission (NDRC) announced that it will accelerate work on a basic, comprehensive artificial intelligence law. Alongside it, safety standards covering the entire AI life cycle are to be created.
The law is also expected to apply to domestic developers such as DeepSeek, Alibaba and Moonshot, which would have to disclose risks and report incidents. No final text exists yet (AI Weekly).
- November 4: At the World Internet Conference summit in Wuzhen, 20 Chinese companies and institutions, including Huawei, Tencent and Alibaba, backed a voluntary pact on the safety of AI agents, including clear limits on permissions and accountability (MLex).
Research
FlashProve: proof of which model answered
On November 2, Anthropic released the research prototype FlashProve as open source (under the Apache 2.0 license). Using zero-knowledge cryptography, it proves that a response was actually generated by a specific model without having to publish its weights, that is, its learned parameters.
Computing the proof is still expensive, roughly 70 to 95 times more demanding on GPUs than ordinary generation, and it works for responses of up to 1,024 tokens. Researchers from EPFL, MIT and the companies StarkWare and Succinct contributed to the project (The Next Web).
Security and incidents
The Nvidia CEO scandal
The founder and CEO of Nvidia, Jensen Huang, admitted to an extramarital affair in an email to employees on Sunday, November 8. This followed reports that he had been targeted by a woman suspected of working with Chinese intelligence.
The same day, the board launched an independent investigation into whether this was a targeted intelligence operation. According to the board, there is so far no indication that company secrets were leaked; three chairs of House committees had already asked Huang for a closed-door briefing on November 4 (South China Morning Post).
- November 2: OpenAI started a security partnership with Cathedral Systems and quadrupled the rewards in its bug bounty program (rewards for reported bugs) to up to $172,000 for a critical finding (AI Weekly).
- November 3: Users pointed out that the Postman desktop app installs plugins into Claude Code and Codex without notice and overwrites their global configuration (AI Weekly).
Sources
- OpenAI: Introducing GPT-6 Cyber
- Google Blog: Gemini Omni 1.2 Flash
- Digit: ChatGPT Images 2.6
- TechCrunch: Claude Personal Finance
- Euronews: Apple begins AI trial in China
- TechNode: Apple confirms Qwen integration
- TechCrunch: Google and K-12 AI controls
- AI Weekly: OpenAI Pando
- TechCrunch: Aina smart ring
- CNBC: Democrats call to rein in AI buildout
- CNBC: AI stocks hammered by midterm shock
- The Next Web: AI stocks sell-off
- Android Headlines: OpenAI Community Compact
- GV Wire / Bloomberg: OpenAI $11 billion funding
- OpenAI: OpenAI to acquire Pierre
- Axios: Meta acquires Conductor
- SiliconANGLE: Meta acquires Conductor
- Crypto Briefing: Anthropic amended IPO filing
- TechBuzz: Anthropic Q3 financials
- WinBuzzer: OpenAI hires Cat Wu
- AI Weekly: China NDRC speeds AI law
- MLex: Huawei, Tencent back AI agent pact
- The Next Web: Anthropic FlashProve
- South China Morning Post: Jensen Huang admits affair
- AI Weekly: OpenAI and Cathedral Systems
- AI Weekly: Postman Desktop plugins