The Day in AI
In short
- Google DeepMind introduced AlphaProof Code and says it formally verified the code of Signal's cryptographic protocol in Lean in four weeks.
- Anthropic launched Claude Meetings, Claude Whiteboards and Claude Sheets in beta, and made Claude Data generally available for warehouse and CRM data.
- OpenAI launched Businesses in ChatGPT in beta, so US ecommerce and online service firms can claim profiles and run their own Business Agents.
- Meta released Muse Ember on 9 November at $0.10 per million input tokens, and Artificial Analysis scores it close to Claude Haiku 5.5.
- Ant Group's Robbyant open-sourced LingBot-Touch, a tactile model for robot hands pretrained on nearly 5,000 hours of touch data.
The day in brief
Google DeepMind says its AlphaProof Code framework formally verified the implementation of Signal's cryptographic protocol in Lean in four weeks, writing over six million lines of proof.
AlphaProof Code is an agentic framework built on several language models. Human experts and the AI choose which properties of the code to prove, and agents running on Gemini Carbon write the proofs in Lean, a proof language whose checker accepts only proofs that are correct. The agents flagged about 1% of the codebase for human review, and Lean checked the rest. DeepMind reports that this kind of verification has historically taken expert teams years.
As a test, DeepMind reintroduced two real bugs into the code. The system caught both and produced machine-checked proofs that the buggy versions were insecure. DeepMind calls this the first formal verification at this depth of a widely deployed cryptographic protocol's implementation, by any method. It's a paper release, and DeepMind hasn't said whether outside teams will get to run the framework. The agents use the model family behind Gemini 4 Carbon, which Google announced on 28 October with access going first to cyber defenders.
Anthropic followed its Fable 5.2 release of 9 November with a set of office tools. Claude Meetings joins Google Meet calls as a visible participant and drafts plans, builds whiteboards and answers questions while people talk. Claude Sheets lets teams create and co-edit spreadsheets with tabs, formulas and dropdowns, and Claude lists every change it makes. Claude Data, previously called Claude Dashboards, connects to data warehouses and CRMs and answers with dashboards, written data stories or narrated videos. New interactive connectors also open apps such as Adobe, Atlassian and HubSpot inside Claude.
OpenAI launched Businesses in ChatGPT in beta. Companies claim and verify a profile in a Business Hub, add product feeds with prices and availability, and see aggregated figures such as mentions and website clicks. Eligible US firms can build a Business Agent that users chat with from the company's profile, in a labelled conversation kept separate from their own ChatGPT chat. Businesses can also pay to make the agent a destination for ChatGPT Ads, and OpenAI cites 1.2 billion weekly ChatGPT users as the audience.
Also in the news
- Meta released Muse Ember on 9 November at $0.10 per million input tokens and $0.50 per million output tokens, with a 1M-token context window.
- Artificial Analysis scores Muse Ember at 35.8 on its Intelligence Index against 36.2 for Claude Haiku 5.5, at about a third of Haiku's cost per task.
- Ant Group's Robbyant reports that adding LingBot-Touch raised average success by 14 to 26 percentage points for four robot action models on 26 real tasks.
- Google announced TPU 8i Max, which pairs two TPU 8i chips in one package, and TPU 8tx for confidential inference, on 9 November.
- Google DeepMind released Gemini Robotics 2.5 and reports a 30% to 100% higher task success rate than its earlier baselines, depending on task type.
- OpenAI deployed Control Room, which reportedly can stop the output of a research inference job within 35 seconds of a decision to intervene.
- Adobe previewed a local MCP server at MAX that reportedly lets agents from OpenAI, Google and Anthropic edit Photoshop, Illustrator and Premiere files directly.
- Anthropic committed about $10 billion more to Microsoft and Google cloud through 2033 in an updated S-1, according to The Information.
- AI Futures Project now expects a superhuman AI researcher in 2027, after METR's report on Anthropic's research and development work.