Afternoon edition — Tuesday 10 November 2026

Google DeepMind introduced Gemini Robotics 2.5, which it calls its most capable vision-language-action model to date: a model that turns camera input and instructions into robot movements. It is rolling out to DeepMind's robotics partners, not as a public API.

DeepMind says it evaluated the model on more than 200 tasks, such as serving food with tongs, tying a bow and pipetting liquid, and that it reaches a 30% to 100% higher success rate than earlier baselines, depending on the task, while finishing tasks 20% to 30% faster.

In a proof of concept with Apptronik and Mercedes-Benz, Apptronik's Apollo 2 humanoid sorted metal brackets from bulk bins into sequencing trays on production lines in Kecskemét, Hungary, and Sindelfingen, Germany, with a 96% success rate, according to DeepMind. The company also says new safety guidance raised safe task completion around obstacles by 33% and around nearby people by 43%.

At its MAX conference in Miami Beach, Adobe previewed local MCP integrations for Creative Cloud that will let users direct their own third-party AI agents to work in Photoshop, Illustrator, Premiere and After Effects, making changes directly in native project files on the user's computer. Layers, vectors, bins and sequences stay editable.

The Verge names agents from OpenAI, Google and Anthropic as examples and reports that a private beta is due in early 2027. Adobe also expanded its own Creative Agent across Creative Cloud apps in beta, previewed always-on agents that run recurring tasks in the background, starting with Firefly AI Assistant, and added a Plan Mode that lets users approve an agent's plan before it runs.

Microsoft told administrators that from 9 December 2026, Claude models will be enabled by default in Microsoft Copilot for UK tenants that have not made a selection, in line with other regions, according to its Message Center notice MC1546435 as reported by The Register. Admins can opt out before then or turn Claude off at any time.

Anthropic works as a Microsoft subprocessor and processes the data in the United States, and UK tenants are not covered by the EU Data Boundary. Microsoft is also retiring the option that allowed Claude only in Copilot in Microsoft 365 apps; tenants using it move to all users. EU and EFTA tenants keep Claude off by default, per Microsoft's documentation.

Google DeepMind presented AlphaProof Code, a framework in which human experts decide what to prove and AI agents write the proofs. Researchers used it to prove key security properties of the compiled code of the Signal protocol, which encrypts messages in Signal and Google Messages, not just of its design.

Agents running on Gemini Carbon generated over six million lines of Lean proofs, and the agents flagged about one percent of the work for human review, DeepMind says. The verification took four weeks. As a check, the team reintroduced two real, previously fixed bugs into an isolated copy of the code, and the system produced machine-checked proofs that the buggy versions were insecure. The Register reports the work covered Signal's post-quantum PQXDH handshake and its Triple Ratchet.

DeepMind says this is the first time the implementation of a widely deployed protocol like this has been formally verified by any method. Signal's chief technology officer, Ehren Kret, is quoted in the announcement calling the approach a major breakthrough. A technical paper accompanies the post.

Google Cloud unveiled Gemini for healthcare, an agentic product built on its Gemini agent that connects to electronic health records, billing platforms, clinical guidelines and Centers for Medicare and Medicaid Services datasets, and runs multi-step workflows such as medical records synthesis and prior authorisation review, with human oversight.

It is available in preview. Launch customers named by Google include Blue Cross Blue Shield of Massachusetts, Hartford HealthCare, HCA Healthcare, Highmark Health, Northwell Health, Optum and Stanford School of Medicine.

Microsoft has been pressing the Trump administration to drop plans to bar imports of new Chinese optical transceivers, the parts that move data over fibre inside AI data centres, Reuters reported on 10 November, citing three people familiar with the matter. The Federal Communications Commission is working on the ban, over concerns the parts could be used to steal data or disrupt US data centres.

Microsoft is one of several companies and trade groups arguing that a ban would slow the data-centre build-out and hurt the US lead over China in AI, Reuters says. Two sources said a threshold covering only 3.2 terabit-per-second transceivers, not expected in volume until 2028, is under consideration.