AI progress — 11 November 2026

Google DeepMind made Gemini 4 Carbon generally available, and GitHub added it to Copilot the same day, so it can be tested in real workflows right away. Alongside it, DeepMind introduced AlphaProof Code for software verification, a claim that still needs outside testing. On the policy and security side, New York’s AI safety law took effect and a vendor breach at Fireworks AI reached Notion customers. Release candidates were otherwise thin today; source summaries were empty, so items rest on headlines.

New models and papers

• Gemini 4 Carbon is now available — Google DeepMind announces general availability of Gemini 4 Carbon. Why it matters: It is a new model tier to evaluate against whatever you run now, and it already shows up in other products (see Copilot below).
• Introducing AlphaProof Code — DeepMind presents AlphaProof Code, which it describes as a way to verify complex software at depth. Why it matters: If the claim holds up, formal verification of real software becomes a tool to try rather than a research topic; the post is DeepMind’s own claim and is not independently tested.
• Kimi K3.1 Pro on Amazon Bedrock — Moonshot’s Kimi K3.1 Pro is now offered through Bedrock; a Kimi-K3.1-Pro repository is also trending on Hugging Face. Why it matters: Teams on AWS can call the model through a managed endpoint without hosting it themselves.

New tools and software

• Gemini 4 Carbon in GitHub Copilot — GitHub adds Gemini 4 Carbon to Copilot. Why it matters: Copilot users can try the new model in their editor today and compare it with their current default.
• modern-web-guidance — A GoogleChrome repository, shown on Hacker News, meant to steer coding agents toward modern web primitives instead of ~2019-era JavaScript. Why it matters: Agent-written front-end code tends to follow older patterns; this is a ready-made guidance file to test against that.
• Anthropic’s AI-powered office suite — The Verge reports Anthropic now has its own office suite. Why it matters: It moves Anthropic from model supplier to a direct competitor in productivity software; check the article for availability details.

Overall digest

• New York’s AI safety law takes effect — TechCrunch reports OpenAI and Anthropic now face new oversight under the law. Why it matters: It is a binding state-level obligation for frontier developers, not a proposal.
• Fireworks AI security breach impacts Notion — Notion’s notice describes a security incident at a third-party AI vendor that affected Notion and its customers. Why it matters: It shows that data sent to an inference vendor is exposed to that vendor’s security; review which vendors see your data.
• Grok Build coding agent leaks confidential info — Ars Technica reports the Grok Build coding agent leaked a large amount of confidential information, some belonging to SpaceXSI. Why it matters: Anyone letting a coding agent touch private repositories should check what the agent can send out.