AI News, Oct 12, 2026: Meta launches Muse Wonder and Muse Wonder Full AI models

Meta returns to the frontier with new models while Microsoft pushes its own chips and governments debate AI oversight.

1. 01
   Meta's Muse Wonder ranks top three on six of ten benchmarksThe Decoder

   Meta's Muse Wonder models placed in the top three on six of ten benchmarks, with the largest gains in agent tasks, though cost per task has nearly tripled.

   Why it matters

   The model is strong at multi-step tasks, but the cost per task has nearly tripled, which could mean higher prices for apps built on it.
2. 02
   Meta launches Muse Wonder and Muse Wonder Full AI modelsMeta AI

   Meta announced Muse Wonder and Muse Wonder Full, its most capable models so far, built for personal and professional tasks where AI acts on a user's behalf.

   Why it matters

   Meta's AI assistants and agent tools may get more capable, and a new top-tier option now competes with other leading AI models.
3. 03
   Microsoft study: clearing chat history helps AI agents find better answersThe Decoder

   Microsoft researchers found AI agents get stuck in limited solution areas as chats grow, and their Fork-and-Flush method clears history, improving results by up to 66 percent.

   Why it matters

   It suggests AI agents working on long tasks may give better results when their conversation history is reset, which could improve tools you use.
4. 04
   White House restores UK access to early AI models for safety testingThe Decoder

   The White House reversed its September order and restored the UK AI Security Institute's early access to models from Anthropic, Google DeepMind, and OpenAI.

   Why it matters

   UK safety testers can again check new AI systems from major labs before release, which affects how well risks are caught early.
5. 05
   Microsoft announces Maia 300, its third-generation in-house AI chipThe Verge AI

   Microsoft unveiled Maia 300, designed to be the primary system for running AI in its products, and plans to deploy Maia chips at multi-gigawatt scale from 2027.

   Why it matters

   Cheaper, more efficient chips could lower the cost of running AI in Microsoft products and reduce reliance on Nvidia.
6. 06
   DeepMind's AlphaProof Nexus solves dozens of open math problemsThe Decoder

   Google DeepMind's AlphaProof Nexus solved dozens of open math problems, including nine Erdős problems, in a peer-reviewed study published in Science.

   Why it matters

   It shows AI can produce checked, new math results, not just answer known questions, which may speed up scientific research.
7. 07
   Microsoft president says AI is less regulated than milkThe Guardian AI

   Microsoft president Brad Smith said AI faces less oversight than milk and washing machines, and called for tougher rules including kill switches and legal duties to victims.

   Why it matters

   A major AI company is calling for stricter rules, which could shape future laws on AI safety and who is liable when AI agents cause harm.