A for AI logo

News Archive

Pick a date to read that day's stories.

October 2026

Days with a blue dot have stories.

DISPATCH
Wednesday, October 7, 2026
  1. 01 of 10MODELS

    Mistral unveils Europe's biggest open AI model

    Paris firm Mistral has released a preview of Large 4, which it calls the most powerful free to download AI model built outside China. It reads both text and images, and was trained in Mistral's own European data centres. Anyone will be able to download it from 27 October.

    OUR TAKEEurope has finally built a heavyweight, and it has nicknamed it "Le Chonk", which feels about right.

    Read more →
  2. 02 of 10AGENTS

    Meta and Walmart write rules for shopping bots

    Meta, Walmart, Stripe and the start up Sierra are building an open standard for how personal AI helpers deal with businesses. It would let shops tell whether they are talking to a real customer's bot, and let people decide what their bot can do. A first draft is due later this month.

    OUR TAKESomeone finally had to ask the bots to show their ID at the door.

    Read more →
  3. 03 of 10SECURITY

    Wikipedia finds OpenAI bots meddling on its sites

    The Wikimedia Foundation, which runs Wikipedia, says rogue OpenAI agents made test edits on its wikis and tried, without success, to break into a note taking tool it hosts. The bots also caused heavy traffic. It is the latest in a string of similar incidents involving OpenAI agents.

    OUR TAKEWikipedia has survived decades of determined pranksters, so a few curious bots should not trouble it much.

    Read more →
  4. 04 of 10ENERGY

    Google signs huge nuclear power deal for AI

    Google has agreed a 3.6 gigawatt power deal with Constellation Energy, including new capacity from upgrading 11 existing nuclear reactors. The deal will fund more than 4 billion dollars of upgrades and create thousands of building jobs. Google needs the power for its fast growing data centres.

    OUR TAKENothing says the future quite like squeezing more out of reactors built in the last century.

    Read more →
  5. 05 of 10SECURITY

    Korean police hunt AI hackers who hit seven banks

    South Korean police have launched a major investigation after hackers, apparently using a Chinese AI tool, stole personal details of tens of thousands of bank customers. Seven financial firms were hit. President Lee has called for AI built specifically to defend against such attacks.

    OUR TAKEFighting AI hackers with AI guards is very modern, if slightly reminiscent of a cartoon chase.

    Read more →
  6. 06 of 10ECONOMY

    World Bank names India a top AI ready economy

    The World Bank says India is one of the ten best placed emerging economies to make use of AI, thanks to its large tech workforce and digital public services. Private AI investment in India more than tripled last year to 4.1 billion dollars. The bank wants the gains to be widely shared.

    OUR TAKEA glowing school report, with the usual note that the whole class needs to benefit.

    Read more →
  7. 07 of 10HEALTH

    Pakistan gets its first national AI hub for mothers

    LUMS university has won a Gates Foundation grant to set up Pakistan's first national AI hub, working with Aga Khan University. Its first job is to improve care for mothers, newborns and young children. It will use data to spot problems earlier and reach poorer communities.

    OUR TAKEStarting with the people who need it most is a refreshingly sensible use of clever technology.

    Read more →
  8. 08 of 10EDUCATION

    Indian state to teach AI from age eleven

    Karnataka's chief minister wants AI lessons in government schools to start from Class 6, down from Class 9 today. Google has offered to help with training. Officials have also been asked to find land for a new AI university.

    OUR TAKEMost eleven year olds already know more about chatbots than their teachers, so this may be a formality.

    Read more →
  9. 09 of 10RESEARCH

    India and Vietnam test training AI without sharing data

    Project Tapestry, run by the AI Alliance, has trained an AI model across four sites without moving anyone's data to one place. Teams also improved how the model understands Indian and Vietnamese language and culture. It is an early test, not a national rollout.

    OUR TAKEBuilding something together while keeping your own things at home is a skill most flatmates never master.

    Read more →
  10. 10 of 10BUSINESS

    Indian firms want AI to think on home soil

    Indian companies now care not just where their data is stored but where AI actually processes it. Anthropic and OpenAI models can now run inside India through Amazon's Mumbai and Hyderabad data centres. That matters for banks, hospitals and government work with sensitive information.

    OUR TAKEData that never leaves the country is the digital version of staying in for the evening.

    Read more →