Skip to content
Newsheir
AILaunch

Google launches Gemini 4 Carbon, an AI model built to find and fix security flaws

Access is limited to vetted cyber defenders for now. Google is also shutting down its Jules coding agent and folding it into its Antigravity platform.

By Newsheir Staff, Newsroom

· 1 min read

ShareXLinkedInBlueskyEmail
In this article
  1. Restricted access
  2. What Google claims
  3. Why it matters
  4. Goodbye, Jules

Google has launched Gemini 4 Carbon, a new artificial intelligence model designed to find software vulnerabilities, test whether they can be exploited and write fixes.

Announced on October 28, Carbon follows Gemini 4 Argon, released two weeks earlier. Google calls it its strongest model, with leading capabilities in cybersecurity and in long tasks that can run for days.

Restricted access

Because a model that can find and exploit security flaws could also be misused by attackers, Google is limiting access. Carbon is first being made available to a group of trusted cyber defenders through its Fairwind programme. Google says it will expand access as soon as possible.

What Google claims

Google reports that Carbon scored 88.2 percent on a benchmark for discovering vulnerabilities, 54.8 percent on one for exploiting them, and 71 percent on one for patching them. The company says it outperforms its previous models but did not name the rival models it compared against, which makes independent comparison difficult.

Why it matters

Security teams struggle to keep up with the number of flaws in modern software, and attackers increasingly use automation. AI systems that can find and fix bugs at scale could help defenders, but they could equally help criminals and state hackers if they became widely available. That tension is pushing AI companies, including Google, OpenAI and Anthropic, towards restricted, "trusted access" releases for their most capable security and biology models.

Goodbye, Jules

Separately, Google is shutting down Jules, its coding agent that works on programming tasks in the background. Jules becomes read-only after November 11 and will go offline on December 31. Its technology will be folded into Antigravity, Google's unified platform for AI-assisted software development.

Code written by Jules is safe because it already lives in users' GitHub repositories, but settings, task schedules and session history need to be exported before the shutdown. Google says the decision is part of an effort to unify its AI coding products, not the result of technical or security problems.

Sources and references

  1. 1.Gemini 4 Carbon: our next frontier model — GooglePrimary source
  2. 2.Google Touts Powerful Cybersecurity Capabilities in Gemini 4 Carbon Model — SecurityWeek
  3. 3.Google says Gemini 4 Carbon beats its predecessors at finding and fixing security flaws — Help Net Security
  4. 4.Google Is Shutting Down AI Coding Agent Jules As It Moves Its Work To Antigravity — Times of AI

AI disclosure This article was drafted with the help of AI tools and was reviewed, fact-checked and edited by our editors before publication. See our AI policy for details. How we use AI.

Spotted an error? Request a correction. Read our editorial standards.

Related coverage

Format: Launch