AI Daily Grind // AI news, freshly ground daily
Cover illustration for Wednesday, November 11, 2026
Morning Token 04 🚀 Launch

Gemini 4 Carbon opens to all, and AI helps verify Signal's encryption

Two weeks ago, Google gave Gemini 4 Carbon only to a select group of 'cybersecurity defenders'. Now everyone can use it. Google calls it its strongest model yet, able to compete with the best from Anthropic and OpenAI. It comes with an agent 'swarm' feature, meaning many agents working together.

Simon Willison notes the launch price: $2 per million input tokens and $10 per million output tokens. A token is a small piece of text. On December 1, the price doubles to $4 and $20. The launch got 287 points on Hacker News.

Google DeepMind also showed AlphaProof Code. Its agents wrote three million lines of proofs in Lean, a language that lets a computer check math, to verify Signal's core encryption. And on Hacker News (429 points), the security firm VulnCheck published a November update tracking vulnerabilities discovered by Anthropic.

Why it matters

Good news first: AI that proves code is correct is exactly the kind of power we want. And still, models that find bugs can find them for attackers too. That is why Google gave defenders a head start, and that head start lasted two weeks.

More on: 🚀 LaunchGeminiGoogleCybersecurity

  1. 01 Anthropic sets its IPO range: up to $1.404 trillion, about $250M per founder Business
  2. 02 New York's AI safety law takes effect: register, disclose, report incidents Policy
  3. 03 35,000 apps built with Grok Build were left wide open Safety
  4. 05 Fewer than 10 neurons: a reported backdoor shows how small edits can hijack an agent Safety
  5. 06 Is AI making our code fragile, and stuck in 2019? Community
  6. 07 Apple is reportedly making an app to run AI models on your Mac Tool