AI agents check Signal's code with proofs

People choose what to prove, agents write the proofs

On 10 November 2026 Google DeepMind presented AlphaProof Code, a system in which people decide what to prove and AI agents do the proving. Experts first turned the security goals of the Signal protocol into formal specifications and reviewed each one; agents running on Gemini Carbon then proved them against the compiled code that runs on devices, not only against the design. The work produced over six million lines of Lean, which checks every proof by machine, and took four weeks. As a test, the team put two old, already fixed bugs back into an isolated copy of the code, and in both cases the system produced a proof that the buggy version was insecure. DeepMind is publishing the verification artifacts and a technical paper.

Source: Google DeepMind: Introducing AlphaProof Code

In brief

Claude Fable 5.2: faster, and able to lead other agents

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026. They are the same underlying model: Fable has stronger safeguards and is open to everyone, while Mythos goes to verified cybersecurity and life-science organisations. Anthropic reports gains in coding, reasoning and vision, and output up to 30% faster than version 5.1. Fable can also coordinate other agents; in one experiment it led over 1,000 subagents for eight days to build a prototype operating system.

Source: Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2

Gemini 4 Carbon opens beyond security teams

Google introduced Gemini 4 Carbon on 28 October 2026 as its strongest model, aimed at cybersecurity and long, multi-step work, and first gave it only to a group of cyber defenders. On 10 November Google made it available to developers, businesses and consumers, among other places through the Gemini API in Google AI Studio and in Android Studio. Google says internal teams have seen it run for days on engineering tasks such as finding and patching bugs in large codebases.

Sources: Google: Gemini 4 Carbon is now available · Google: Gemini 4 Carbon, our next frontier model

Mistral publishes the weights of Mistral Large 4

On 29 October 2026 Mistral released the weights of Mistral Large 4, which it calls "Le Chonk": a model with about one trillion parameters, of which about 52 billion are active per token, and a context of one million tokens. Mistral aims it at agentic coding and defensive cybersecurity. The licence is a modified MIT licence that allows commercial use; offering the model as a hosted service needs a separate licence. The weights are on Hugging Face.

Sources: Mistral AI: Le Chonk, built for coders, trusted by defenders · Hugging Face: Mistral-Large-4-1T-A52B

DeepSeek shares a coding agent that waits to be asked

DeepSeek published dsh-superagent on GitHub on 31 October 2026, under the MIT licence. It watches one GitHub repository, and when an allowed person asks for work, it opens a work item, does the work in its own session and answers in the same conversation, for example with a pull request. Its rule is to take as little human attention as possible. The README calls it a research prototype, not an official DeepSeek product.

Source: GitHub: deepseek-ai/dsh-superagent-prototype

Agent Skills gets a neutral home

On 22 October 2026 Anthropic gave the Agent Skills standard, the format in which a skill is a folder with a SKILL.md file, to the Agentic AI Foundation under the Linux Foundation. The foundation already holds the Model Context Protocol. The planned roadmap includes versioned releases, a formal proposal process and security guidance for people who write and install skills. Anyone can take part without joining the foundation.

Source: Anthropic: Donating Agent Skills to the Agentic AI Foundation

Ask GitHub from Slack and Teams

Since 6 November 2026 @GitHub in Slack and Microsoft Teams is generally available. From a conversation you can investigate a bug, create an issue, make changes and open a pull request. Issues and pull requests it creates say that they came from such a conversation and who asked.

Source: GitHub Changelog: @GitHub in Slack and Teams is generally available

EU asks how elections should deal with generative AI

On 10 November 2026 the European Commission opened a consultation on revised election guidelines under the Digital Services Act for the largest platforms and search engines. It names the fast change of technology, "particularly generative Artificial Intelligence", as one reason for the update. Views can be sent until 23 November 2026.

Source: European Commission: draft revised election guidelines under the DSA

From Skribly

The Skribly Discord server is open

The short address dc.skribly.pl takes you to the invite of the Skribly Discord server. It is the place to ask questions and talk with the people who build Skribly.

Source: dc.skribly.pl

One page for all our projects

projects.skribly.pl lists the projects our team builds and will build, each with what it is, what it is made of and where it stands. Skribly is the first; more will follow.

Source: projects.skribly.pl

Docs and the UI kit are online

docs.skribly.pl holds the guides and reference for every Skribly tool and now has search on its home page. ui.skribly.pl shows the design system live: tokens, plain CSS and React components on React Aria, in light, dark and high contrast.

Sources: docs.skribly.pl · ui.skribly.pl

Skribly Agent will be free to download

Skribly Agent, the desktop app where coding agents work side by side, will be free to download and use. There is no date yet: the installers will appear on the download page when the app is published. Skribly is a non-commercial project we build in our spare time.

Source: skribly.pl/download

One thing to try

Look inside an MCP server with MCP Inspector

MCP Inspector is the free, open-source tool of the Model Context Protocol project for checking an MCP server. Run npx @modelcontextprotocol/inspector to open its web view, connect it to a server you already use and see its tools, resources and prompts. Prefer the terminal? npx @modelcontextprotocol/inspector --cli node build/index.js --method tools/list lists the tools of a local server. Version 2 stores tokens and secrets in the system keychain where there is one, so read its note on secrets first.

Source: GitHub: modelcontextprotocol/inspector

Read the whole issue