Google releases Gemini 4 Carbon first to a subset of cyber defenders

For the CISO of a regional US bank

Google announced Gemini 4 Carbon, its next frontier model, on Oct. 28 and said it will make the model available starting this week to a subset of cyber defenders.1

Google said it is taking a deliberate approach to rollout, is engaged in the U.S. government's voluntary process for pre-release model testing, and is gathering feedback and iterating on guardrails before wider release.1 Access will expand gradually to additional trusted partners through the Fairwind Program, which Google DeepMind describes as giving high-priority defenders such as governments, healthcare providers and telecommunications services early access to advanced models.12

Google said Carbon, evaluated without guardrails, scored 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on Collinear's CWE-bench for patching.1

Google said it will release Carbon to developers, businesses and consumers through its product surfaces "as soon as possible" and did not give a date.1

Google said Carbon will launch at the same introductory price as Gemini 4 Argon, $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input token price.1

Google said its teams have seen Carbon run for days on multi-step engineering tasks, and that Argon was launched two weeks before Carbon. The benchmark results are Google's own reporting, and the post did not name an independent verifier for them.1

Why it matters

Google is giving defenders a head start, but the same model skills apply to attack, and the benchmark scores do not show how Carbon performs on your own code. Security teams should plan on the assumption that attackers may gain similar tools, and ask vendors and cloud account teams how to qualify for Fairwind access.

What’s next

Google said it will phase in more Fairwind partners and then open Carbon to developers, businesses and consumers, but gave no date.1

From the brief

Our take. Carbon scores 88.2% at finding flaws and 71.0% at patching them, and the same model scores 54.8% at exploiting them. Defenders get it first; your patch calendar should assume attackers get it next.

Action. Vulnerability management: Shorten critical-patch deadlines for internet-facing systems to match faster AI-found exploits.

What our records say

Gemini Enterprise sells banks a per-seat AI app for employees plus an agent-building platform on Google Cloud, and since August 2026 a financial-services edition for capital markets and corporate banking built with bank design partners; among the 100 banks tracked, Wells Fargo, BNY, Citi, Zions and Goldman Sachs use Google's agent tools or Gemini models.1 It includes finance skills and secure MCP connectors to licensed data providers such as FactSet, LSEG and S&P Global.2

  1. Gemini Enterprise: AI programme summary (part 1)
  2. Gemini Enterprise: Gemini Enterprise for Financial Services

From our records

Sources

  1. Primary sourceIntroducing Gemini 4 Carbon Google
  2. Primary sourceFairwind Program — Google DeepMind Deepmind