Google announcement October 28, 2026; independent verification check October 29, 2026
Gemini 4 Carbon Goes to Cyber Defenders First: Google Strongest Model Has No Public Date and One Independently Confirmed Cyber Score
Google announced Gemini 4 Carbon on October 28 as its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities, and is releasing it first to a subset of cyber defenders, then gradually to vetted partners through its Fairwind Program, before any developer, business or consumer availability, for which Google has given no date. An independent verification review published after the launch separates what can be checked from what rests on Google word. Carbon ranks first on version 5 of the Artificial Analysis Intelligence Index at 59, one point ahead of Claude Opus 5.5, a lead confirmed on the index itself. Of the three headline cyber scores, only one is independently confirmed: 71 percent on CWE-bench, the patching benchmark, confirmed on the Collinear leaderboard dated October 28. The other two are Google self-run figures that no outside party has reproduced: 88.2 percent on SEC-bench Pro, where the official leaderboard best verified entry is 58.4 percent on a different split, and 54.8 percent on ExploitGym, run by Google with mitigations off. Google has published no Carbon model card and no Carbon safety report; the linked safety document is the Argon report, which does not mention Carbon. Access through Fairwind is limited to internal cybersecurity, incident response or penetration testing teams at vetted organizations, behind phishing-resistant multifactor authentication, with no sharing or resale of access. Google stated an introductory price of $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20, although there is no API model ID and the model does not appear on Google pricing page yet. Google also says it is engaged in the United States government voluntary process for pre-release model testing.
Executives
The governance lesson is in the evidence split: the general intelligence lead is independently confirmed and narrow, one of three cyber scores is independently confirmed, and there is no model card or safety report for this specific model yet. A defenders-first rollout with vetted access and no public date is a reasonable control for a model whose exploit score was measured with mitigations off, and it is also a reason to keep your own evaluation evidence, not vendor tables, as the basis for any adoption decision under the NIST AI RMF.
CIO / CISO
If your security team wants Fairwind access, prepare the record the program asks for: named internal team, phishing-resistant MFA, authorized defensive use only, and no redistribution. If you are evaluating Carbon against Argon, GPT-6 Astra or Opus 5.5 for security work, weight the CWE-bench result, the one score an outside leaderboard confirms, and run the SEC-bench and ExploitGym claims through your own authorized test set before treating them as capability. Log the missing model card and safety report as open evidence items in the AI inventory.
Engineers
Plan around absence: there is no API model ID, no Vertex listing and no app availability to integrate against today, so any build work is premature. When access arrives, budget for the operating profile the independent measurements show, long thinking times measured in tens of minutes per hard task, and test patching quality on your own codebase, since the confirmed strength is in the audit and repair stage rather than in unverified exploit claims.
Everyone (non-IT)
A new top-ranked AI model exists that most people cannot use yet, and that is deliberate: Google is giving it first to vetted security teams because the same skill that finds and fixes flaws can be pointed the other way. Nothing changes in the Gemini app or your accounts today, and any message offering early Carbon access outside Google official channels should be treated as a likely scam.