What Is Gemini 4 Carbon? Why Google Is Giving It to Cyber Defenders First

Google unveiled Gemini 4 Carbon on October 28, 2026, touting its vulnerability-hunting skills, but only a subset of cyber defenders can use it for now. Here is what Google has confirmed and what it means for everyone else.
Key Takeaways
- Google unveiled Gemini 4 Carbon on October 28, 2026, calling it its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities.
- Carbon is currently available only to a subset of cyber defenders, with access to be expanded gradually to more trusted partners through the Fairwind Program.
- Google says it will bring Carbon to developers, businesses and consumers as soon as possible, but as of November 3, 2026 it has given no date.
- In tests Google published, run without guardrails, Carbon scored 88.2% on the vulnerability-finding benchmark SEC-bench Pro.
- The Fairwind Program has more than 650 partners, requires phishing-resistant multi-factor authentication and limits use to internal security teams.
Gemini 4 Carbon is a new AI model Google announced on October 28, built around finding and fixing vulnerabilities and working on long tasks. Rather than opening to the public at launch, Carbon is going to a subset of cyber defenders first. This article sums up what Google has officially announced, how the staged rollout works and what it means for ordinary users.
What Is Gemini 4 Carbon?
Carbon is Google's newly announced AI model, which the company calls its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. It arrived about two weeks after Gemini 4 Argon.
The announcement was signed by Koray Kavukcuoglu, Senior Vice President of Google DeepMind and Chief AI Architect at Google. Google DeepMind is Google's AI research division. Long-horizon tasks are multi-step jobs an AI carries out on its own over an extended period; Google says internal teams have seen Carbon work on multi-step engineering tasks for days.
| Item | What Google has announced |
|---|---|
| Announcement date | October 28, 2026 |
| Positioning | Google's strongest model, with state-of-the-art cybersecurity and long-horizon capabilities |
| Who can use it now | A subset of cyber defenders |
| Next step | Gradual access for more trusted partners through the Fairwind Program |
| Developers, businesses, consumers | "As soon as possible," with no date given |
| Pricing (per million tokens) | At launch, $2 per million input tokens and $10 per million output tokens; $4 and $20 after the introductory period |
A token is the unit AI models use to measure text. These prices are the figures Google gave in its announcement as of November 2026; check Google's official notices for current pricing.

How Strong Are Carbon's Security Skills?
In tests Google published, run without the usual guardrails, Carbon scored 88.2% at finding vulnerabilities, 54.8% at exploiting them and 71.0% at patching them, according to the company's own figures.
| Benchmark | What it measures | Carbon's score |
|---|---|---|
| SEC-bench Pro | Finding vulnerabilities | 88.2% |
| ExploitGym | Exploiting vulnerabilities, that is, writing code that can actually attack them | 54.8% |
| CWE-bench | Patching vulnerabilities | 71.0% |
"Without guardrails" means the tests were run without the safety mechanisms normally used to block misuse, so the results are closer to the model's raw capability. All three figures are results Google itself published in its announcement.
Google also says Carbon builds on Argon's safety measures, including model training, real-time misuse defenses, resistance to prompt injection and monitoring of the model's reasoning. Prompt injection means hiding malicious instructions in a web page or document to trick an AI into following them. Google says that on Gray Swan's indirect prompt injection benchmark, Carbon's attack success rate was 0.0%, compared with 0.7% for Argon.
Why Are Cyber Defenders Getting It First?
Google says it is taking "a deliberate approach to rollout": letting some defenders use Carbon first while it gathers feedback and keeps improving its safeguards, then gradually expanding to more trusted partners.
Google's announcement does not explain its reasons in more detail, saying only that "pushing the frontier forward requires moving fast while remaining deliberate about deployment." The Fairwind Program's website, for its part, states that partners may only carry out "dual-use" tasks, and that malicious tasks such as creating malware are not permitted. Dual-use means the same technology can serve defense or be turned to attack: the ability to find vulnerabilities and write exploit code can help defenders verify and fix problems, but attackers could also use it for break-ins.
Google also says it is taking part in the US government's voluntary process for pre-release model testing. Google's roundup of October AI news, published on November 2, likewise says Carbon is in pre-release testing with the US government and a subset of trusted cyber defenders through the Fairwind Program, and that a broader launch to developers, businesses and consumers will follow once safety protections are refined.

What Is the Fairwind Program?
It is a Google DeepMind program that gives vetted defenders such as governments, healthcare providers and telecom services a head start with advanced AI models, so they can rapidly discover and fix critical vulnerabilities.
The Fairwind website currently says partners get access to Gemini 4 Argon; in its Carbon announcement, Google says Carbon will also be opened up gradually through the program. According to the website, Fairwind works under these rules:
- It currently has more than 650 partners worldwide, with priority given to governments and national cyber authorities, critical infrastructure operators and core technology platforms.
- Applicants are vetted, including background checks and a review of their ethical record.
- Participants must agree to user-level authentication, phishing-resistant multi-factor authentication and applicable access controls.
- Access may be granted only to internal cybersecurity, incident response or penetration testing teams, and organizations must track employee access and use.
- Use is limited to defensive or academic work such as authorized threat simulation, reverse engineering and malware analysis. Creating malware is not allowed, and access may not be shared or resold.
Phishing-resistant multi-factor authentication refers to sign-in methods such as passkeys and hardware security keys that cannot be stolen even if you land on a fake site. Anyone can use them on their own important accounts; to learn how they work, see What is a passkey.
What Does This Mean for Ordinary Users?
As AI gets better at finding vulnerabilities, flaws may be found and fixed faster, and software updates may come more often. The most practical step is to turn on automatic updates and not put off update notices.
Two security stories this month show why. The PHP project fixed two critical vulnerabilities on October 21, and the US Cybersecurity and Infrastructure Security Agency (CISA) added them to its Known Exploited Vulnerabilities catalog on October 26; see What is PHP2Shell. On October 29, Synology said attackers were exploiting a vulnerability patched back in December 2024 to encrypt data on NAS devices that had not been updated; see Synology NAS ransomware attacks. In both cases an official fix was already available, and the risk fell on systems that had not been updated.
Here is where to start:
- Turn on automatic updates for your phone, computer and browser, and restart promptly when prompted so updates take effect.
- Keep devices such as your home NAS and router updated too, and don't expose their management interfaces directly to the internet.
- Plan to replace old devices that are no longer supported and no longer receive security updates.
- Switch important accounts such as email and online banking to phishing-resistant sign-in methods like passkeys or security keys.

For when Carbon opens to general users, and for its specifications and pricing, rely on Google's official announcements.
Frequently Asked Questions
Can ordinary users try Gemini 4 Carbon now?
Not yet. Google says Carbon is currently available only to a subset of cyber defenders and will come to developers, businesses and consumers as soon as possible, but as of November 3, 2026 it has not given a date.
How is Gemini 4 Carbon different from Gemini 4 Argon?
Google describes Carbon as a stronger model announced about two weeks after Argon, with particular gains in cybersecurity and long-horizon tasks. Both launched at the same introductory price of $2 per million input tokens.
How can a company apply to join the Fairwind Program?
The Fairwind website says Google Cloud customers can contact their Cloud account team, while others should start from Google Cloud's getting-started page. Applicants are vetted, and no review timeline is given.
Sources: Google: Introducing Gemini 4 Carbon (2026), Google DeepMind: Fairwind Program, Google: The latest AI news we announced in October 2026 (2026)