Google unveils Gemini 4 Carbon for flaw finding, exploitation tests, and patches
Original title: Google Launches Gemini 4 Carbon With State-Of-The-Art Cybersecurity Capabilities
Selected defenders get first access; Google cites 88.2% on SEC-bench Pro and has not set a general-availability date.
Google announced Gemini 4 Carbon on October 28, 2026, as a follow-on to Gemini 4 Argon aimed at multi-day security and coding work. Google says the model finds software flaws, tests whether they can be exploited, and writes root-cause patches. Access starts with selected cyber defenders and later trusted partners in the Fairwind Program, plus voluntary U.S. government testing, not a public launch.
Unconfirmed: Scores, patch-quality gains versus an unnamed rival, and safety metrics are Google’s own claims; no general-availability date is set.
- Google reports 54.8% on ExploitGym and 71.0% on CWE-bench for patching.
- Security benchmarks were run without guardrails, so they may not match deployed services.
- Carbon scored 59 on Artificial Analysis Intelligence Index v5.0.
- Gray Swan indirect prompt-injection attack success was 0.0%, versus Argon’s 0.7%.
- Intro pricing is $2 per million input tokens and $10 per million output tokens.
- Vendors
- Countries
- United States
- AI models
- Gemini 4 CarbonGemini 4 Argon
From the original article
2 min read at cybersecuritynews.com
Google has announced Gemini 4 Carbon, its latest AI model built to find software flaws, test whether attackers could exploit them, and produce fixes. Unveiled on October 28, 2026, Carbon builds on Gemini 4 Argon and targets complex security and coding work that can take days. Access will begin with selected cyber defenders, not a […]
By Abinaya · cybersecuritynews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.