AI Briefing

Google's AI empire had an earthquake this week: Demis Hassabis is handing over the DeepMind wheel, and Jeff Dean is walking out the door after 27 years to build a startup Google is quietly funding. Meanwhile Meta crashed the coding-agent party with Muse Code, Rust told LLMs to keep their hands off pull requests, and OpenAI teased its next model by casually solving ten math problems experts couldn't. The theme of the day: the people and the machines are both reshuffling who's in charge.

01IndustryFortune ↗

Demis Hassabis Steps Down as Google DeepMind CEO, Handing Day-to-Day Control to Kavukcuoglu

Demis Hassabis, who has run DeepMind since 2010, is moving to chairman of Google's AI unit and adding the title of Alphabet chief scientist while continuing to lead drug-discovery spinout Isomorphic Labs. CTO Koray Kavukcuoglu takes over daily operations, reporting straight to Sundar Pichai and overseeing Gemini, frontier research and the developer teams. It's Google's biggest AI reshuffle yet, and it lands as the company scrambles to close the gap with OpenAI and Anthropic.

02IndustryTechCrunch ↗

Jeff Dean Leaves Google After 27 Years to Launch Discovery Loop, and Google Is Backing It

Alphabet chief scientist Jeff Dean is leaving alongside Sanjay Ghemawat, Oriol Vinyals and Quoc Le to co-found Discovery Loop, a public-benefit startup aiming to partially automate the scientific method by running thousands of AI experiments in parallel. Seed funding is led by Radical Ventures and Khosla Ventures, and Google itself is a founding investor, first-year compute supplier and cloud partner. The exit of some of Google's most decorated researchers shows how much elite AI talent is now spinning out to chase bigger swings.

03ProductCNBC ↗

Meta Launches Muse Code, an Agentic Coding Tool Aimed Squarely at Claude Code and Codex

Meta unveiled Muse Code, a terminal-based agentic coding tool that can plan, write and validate code for complex software-engineering tasks, now in beta for macOS and Linux. It ships alongside an upgraded Spark model and undercuts rivals with API pricing of $1.25 per million input tokens and $4.25 per million output tokens. The launch drops Meta directly into the developer-tools brawl against Anthropic's Claude Code and OpenAI's Codex as it hunts for enterprise revenue.

04Open SourceInside Rust Blog ↗

Rust Draws a Line: LLMs Can Read the Code, But Not Write the Pull Requests

Five Rust teams adopted a policy for the rust-lang/rust repo that lets contributors use LLMs to read, analyze and learn from code, but bars AI-generated text in pull requests, issues, documentation and nontrivial source comments. The rules are a direct response to a flood of low-effort AI-authored 'slop' PRs, with only a narrow experimental exception for AI-written code in low-risk areas that both author and reviewer can fully explain. It's one of the clearest stances yet from a major open-source project on where AI belongs in the contribution pipeline.

05ResearchSiliconANGLE ↗

OpenAI's Unreleased Astra Cracks 10 Open Math Problems for About $2,000 in Compute

OpenAI says an internal version of its next major model, Astra, resolved or made substantial progress on ten long-open problems across eight fields of mathematics and theoretical computer science, including the first explicit construction of a non-sofic group. Every result ships with machine-checkable Lean 4 certificates on GitHub so mathematicians can verify the proofs rather than take OpenAI's word, and the whole run cost roughly $2,000 in compute. It's a signal that frontier models may be crossing from acing benchmarks to producing original, verifiable research.