AIIntelGet the Briefing

Latest

BriefingNo. 3

AIIntel Weekly Briefing No. 3

26 October – 2 November 2026

Issued

This week

Official steps this week include an open investigation and a draft rule, beside a publisher control from Microsoft. The UK Competition and Markets Authority published a press release on 30 October saying it has opened an investigation into whether Meta's commercial terms for businesses that use its platforms breach competition law. The release says those terms could let Meta collect content from business and publisher websites and use it in Meta's own products, including AI features. India's telecom regulator, in Press Release No. 145/2026 dated 30 October, published draft guidelines on the use of AI in telecom networks and services and asked for written comments by 27 November. Microsoft's Bing blog, also dated 30 October, says it will offer separate preferences for AI training and for grounding, and that it will start honouring the new preferences as it crawls from 1 February 2027.

Company announcements fill the rest of the issue. Axios reports that Chris Lehane, OpenAI's chief global affairs officer, described plans to press for mandatory limits on the most powerful AI systems and for state legislation on data-centre projects after the US midterms, and that OpenAI will argue for calling new sites "AI compute campuses". A separate Bloomberg account, citing the Financial Times, says OpenAI is now seeking about $20 billion at a $1.25 trillion valuation, down from $30 billion at a $1.4 trillion valuation. Apple's China newsroom says the first Apple Intelligence features for iPhone users in China will open for testing on 4 November and reach users in December with iOS 27.3, with Alibaba's Qwen in some experiences, and without Siri AI. Google's research blog says a Project Suncatcher satellite produced what the post calls the first AI response from a data center class TPU in orbit. Amazon's results release describes a $1 billion community investment in US data-centre communities. Mistral says it released the weights of Mistral Large 4, which it calls a 1-trillion-parameter model. Cursor's notice says a tampered copy of a Claude Code extension was on its marketplace for part of 29 October, and that Anthropic did not develop those versions.

Regulation

AI news

Plus 21 more stories

Tracker changes

5 more in the tracker

Products and releases

  • Apple

    Apple Intelligence (China)

    First core Apple Intelligence features for iPhone users in China, announced 31 October 2026. Developer beta from 4 November 2026; the release to users in…

    Source (opens in a new tab)

  • Mistral

    Mistral Large 4 (open weights)

    Weights released 29 October 2026 for the model that was in public preview from 6 October. Mistral's post calls it a 1-trillion-parameter multimodal model. The…

    Source (opens in a new tab)

  • NVIDIA

    Nemotron 3.5 Transcribe

    Open-weight speech model introduced 28 October 2026. NVIDIA's blog says it transcribes 40 languages, including seven Indic languages, with about 4.3 billion…

    Source (opens in a new tab)

  • Google

    Gemini 4 Carbon

    Frontier model announced 28 October 2026. Google says it is available first to a subset of cyber defenders, then to further trusted partners through the…

    Source (opens in a new tab)

  • Alibaba

    Qwen-Audio 3.1 Realtime Flash

    Listed on Alibaba Cloud Model Studio's model-release table for 27 October 2026, international service, model id qwen-audio-3.1-realtime-flash. The table calls…

    Source (opens in a new tab)

  • Google

    Gemini Robotics Embodied Reasoning 2

    Public preview on the Gemini platform, announced 27 October 2026. Google Cloud describes it as a specialised Gemini model for high-level robot planning,…

    Source (opens in a new tab)

  • Google

    Genie 4

    World model announced on 27 October 2026. Google says a user can steer a scene in real time with natural language and that each world generates native audio,…

    Source (opens in a new tab)

  • OpenAI

    GPT-6.1 Terra

    AWS posted on 26 October 2026 that the model is generally available on Amazon Bedrock, with text and image input and a context window of up to 1 million input…

    Source (opens in a new tab)

Coming up