AIIntelGet the Briefing

Latest

NewsletterNo. 2

AIIntel Weekly Briefing No. 2

19 – 26 October 2026

Issued

This week

Governments account for much of this week. HM Treasury published a page on 24 October that lists an AI biosecurity lab and an additional £120 million for the AI Security Institute, and that also says a major package will be announced by the Chancellor next week. SVT reports that Sweden's Government Offices will block the browser versions of ChatGPT and Claude on staff work computers from 26 October, while staff keep access to those models through the offices' Intric platform. Canada's privacy commissioner published comments on a government paper about AI transparency. The Congressional Record for 22 October lists H.R. 10873, a bill for the National Institute of Standards and Technology to carry out activities on certain AI attacks; CyberScoop reports that the bill, which is not law, is aimed at distillation attacks. Japan's Digital Agency says a Cabinet decision on 20 October set up an AI and Digital Reform Promotion Council, which held its first meeting on 22 October. Pennsylvania's governor approved the Artificial Intelligence Deceptive Advertising Act on 22 October, as Act No. 99 of 2026.

Company announcements sit beside that official warning from the UN's telecoms agency. OpenAI said it has signed an agreement with Poland's NASK National Research Institute to explore joint work on AI safety, and, in a separate post, that its frontier models and Codex workflows are coming to Google Cloud's Model Garden, with general availability planned in the coming weeks and Microsoft remaining its primary cloud partner. Kimi introduced K3.1 Pro, which it calls a 3T-class model with a 1-million-token context window, and said full weights will be released by 4 November. Al Jazeera reported that Tomas Lamanauskas, deputy secretary-general of the International Telecommunication Union, warned that unequal access to AI could widen the gap between countries. He said Africa and Latin America together account for 3.3 percent of global data-centre capacity, and that 2.2 billion people remain without internet access.

Regulation

AI news

Plus 5 more stories

Tracker changes

3 more in the tracker

Products and releases

  • Kimi

    Kimi K3.1 Pro

    First model in the K3.1 series, described as a 3T-class open-weight model with a 1-million-token context window, plus Kimi K3.1 Pro Highspeed at 2-3× the…

    Source (opens in a new tab)

  • NVIDIA

    Nemotron 3.5 Bolt E4B 4GB

    Open small language model for local conversational agents, published 23 October 2026. Built on Gemma 4 E4B with Nemotron post-training, for voice and device…

    Source (opens in a new tab)

  • OpenAI

    Decisions API

    Generally available on 22 October 2026, built with GPT-6 Luna, for questions with bounded answers (a probability, a choice, or a score). OpenAI says it is up…

    Source (opens in a new tab)

  • Mistral

    Voxtral 3

    Speech-to-text and text-to-speech family plus a voice-agent SDK. The announcement page is timestamped 22 October 2026, 09:00 (UTC+2); Mistral's docs changelog…

    Source (opens in a new tab)

  • MiniMax

    MiniMax M3.1 Flash

    Next-generation Flash model for agentic tasks, coding, computer use and cybersecurity. MiniMax says online inference is nearly three times faster, and that…

    Source (opens in a new tab)

Coming up