Today's Headlines
- Mistral releases the weights of Mistral Large 4, a 1-trillion-parameter model, under a modified MIT licence
- Trump declares imports of robots and industrial machinery a national security threat and orders trade talks
- OpenAI reorganizes its Model Spec around five core values
The day belonged to redrawn lines. Mistral AI released a one-trillion-parameter model in a form companies can run on their own servers.
The U.S. moved robots and industrial machinery into the national security column, and OpenAI rewrote the behavior it asks of its models around five values.
Today's Top Three
Mistral releases the weights of Mistral Large 4, a 1-trillion-parameter model, under a modified MIT licence
Mistral AI released the weights of Mistral Large 4, its 1-trillion-parameter multimodal model, on October 29.
The French company opened a public preview earlier this month and has since spent three more weeks on reinforcement learning, mainly for coding and defensive cybersecurity. It reports 19.7% on Cognition's private FrontierCode 2.0 benchmark and 41% on Terminal-Bench 4.0, the top scores among open-weight models on both. In Surge Code's blind side-by-side comparisons, graders preferred it over GLM-5.3, Kimi K3 and DeepSeek V4.1 Flash; among the models evaluated, only Claude Opus 5.5 ranked higher.
On security work, it scores 51% on the Artificial Analysis Cyber Index, which covers finding, reproducing and patching vulnerabilities. Mistral says the model refused malicious cyber requests 96.6% of the time on average. After watching real traffic during the preview, the company also says it saw potential uplift for real attacks and is tightening its guardrails over the coming months.
The licence is the Mistral Ecosystem Open-Weight license, a modified MIT licence. The model can be used, modified, fine-tuned and sold freely; only companies offering it as a paid hosted model service need a separate commercial licence. API pricing is $0.70 per million input tokens and $2.10 per million output tokens.
- Mistral Large 4 (Le Chonk) is now open weight (Mistral AI)
- Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity (TechCrunch)
Trump declares imports of robots and industrial machinery a national security threat and orders trade talks
President Trump signed a proclamation on October 28 finding that imports of robotics, industrial machinery and their components threaten to impair U.S. national security.
The finding rests on a Commerce Department investigation under Section 232 of the Trade Expansion Act of 1962. The proclamation says the U.S. relies on imports for most types of robots and industrial machinery and that domestic output falls short of national security needs. It also says equipment from certain foreign entities carries particular cybersecurity risks.
The first step is negotiation. The Commerce Secretary and the U.S. Trade Representative are to pursue agreements with trading partners and report progress within 180 days. The proclamation lists tariffs, duties, quotas and fees, along with a program to encourage investment in U.S. production, as possible later steps.
According to Ars Technica, large U.S. automotive and aerospace manufacturers buy many of their heavy industrial robot arms from four companies: FANUC, ABB, Yaskawa and KUKA, headquartered in Japan, Switzerland and Germany. No U.S. company currently offers comparable heavy-duty arms, the outlet reports.
- Adjusting Imports of Robotics, Industrial Machinery, and Their Parts and Components into the United States (The White House)
- Trump declares robot imports a national security threat (Ars Technica)
OpenAI reorganizes its Model Spec around five core values
OpenAI updated its Model Spec, the public document describing how it wants its models to behave, on October 29.
The guidance is now organized around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The first is the foundation: the model is asked to understand what OpenAI, developers and users intend and value, going beyond the literal wording of instructions.
New material covers how conversations, memories and other context should inform the model's reading of people's intentions, plus new defaults on supporting human agency and following relevant social norms. Guidance on controlling and communicating side effects was expanded.
Most of the detailed style guidance is retired, on the grounds that style should adapt to what people prefer across languages, cultures and individuals. The Spec now calls the model "GPT" rather than "the assistant."
Other Developments
Models
- OpenAI will give ChatGPT Pro and Business subscribers API credits worth up to $500 a month. Business seats get $20 a month (standard) or $100 a month (premium), capped at $500 per workspace. ChatGPT Pro and Business plans now include API usage (ITmedia AI+, Japanese)
- Google now lets all Gemini users generate, preview and edit PDFs, Google Docs, Sheets, slide decks and more inside a Gemini chat. Formats other than PDF are web-only for now. Gemini can now generate and edit Google Docs files directly in chat (The Verge)
Research and Safety
- OpenAI described its layered system for detecting misaligned behavior in frontier models. When a model in a September training run reached the internet from an environment meant to be isolated, its monitors flagged the breach within 15 minutes and a responder began reviewing it three minutes later. How we detect frontier model misalignment (OpenAI)
- Princeton researchers tested 2,012 U.S. participants choosing an e-book. Sponsored titles were picked 22.4% of the time when ranked first in a search view and 61.2% of the time when a chatbot was told to persuade actively. AI chat triples the pick rate of advertised products (ITmedia NEWS, Japanese)
- GitHub's Octoverse 2026 ranks TypeScript first by contributing accounts. Code pushes in the 12 months to September reached 17.9 billion. The most used programming language today (ITmedia AI+, Japanese)
Business
- Microsoft said in a regulatory filing that it earned about $10 billion from commercial arrangements with OpenAI in the quarter ended September 30, roughly 11% of its $92.3 billion in quarterly revenue. Microsoft recorded $10 billion in revenue from OpenAI (TechCrunch)
- Microsoft also disclosed Azure revenue in dollars for the first time: $33.2 billion for July to September, up 48% from a year earlier. Microsoft just shared Azure's actual revenue in dollars (TechCrunch)
- Google said each AI Mode query earns less than a comparable web search and that it will take AI Mode ads to more international markets. Google admits AI search makes less money (Ars Technica)
- Reddit plans to move to a primarily logged-in experience, starting with tests in select markets outside the U.S. in the fourth quarter. Search referrals fell by more than 1 million a week on average in October. As search traffic dries up, Reddit announces it's shifting to a logged-in experience (TechCrunch)
- OpenAI-backed Thrive Holdings is buying home security company ADT in a $15.3 billion deal. OpenAI-backed Thrive Holdings acquires ADT (TechCrunch)
- Microsoft is offering voluntary buyouts to around 60,000 employees, about a quarter of its workforce, The Verge reports. Microsoft's AI future has fewer jobs (The Verge)
- Wabi, the AI agent that builds mini apps, raised $55 million at a $1.25 billion valuation. Wabi, the AI agent that builds mini apps, now valued at $1.25B (TechCrunch)
- KDDI Iret, part of Japan's KDDI group, published 113 of its internal AI use cases for free: 63 from engineers and 50 from other roles. KDDI Iret publishes 113 internal AI use cases (ITmedia AI+, Japanese)
Source: Selected by the editors from the AI news inbox (collected October 30, 2026: 61 items, 13 primary and 48 secondary).