Google Launches Gemini 4 Argon With Guardrail

Google on Wednesday announced Gemini 4 Argon, its new frontier AI model, which it is initially rolling out to a select group of trusted cyber defenders through its Fairwind Program.

Google’s internal teams are also using the model. The company says it will expand access gradually as it gathers feedback from early testers.

The tech giant says Argon is designed for complex workflows in software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense. On the security side, the company says it trained the model to be highly capable at cyber defense, and that it can autonomously find, validate, and patch critical software vulnerabilities.

“For trusted defenders and our own internal teams at Google, we’ll be releasing Argon without cyber guardrails so they can leverage its full frontier-level cybersecurity defense capabilities,” the company said.

Google launched Fairwind in early September as a limited access program for governments, Google Cloud customers, and cybersecurity partners. It initially combined the Gemini 3.8 Flash Cyber model with Google’s CodeMender harness, which finds, verifies, and fixes vulnerabilities. At launch, the program had more than 650 participating partners.

Google says Argon found a flaw in hospital software

Wiz, which Google acquired earlier this year, is using Argon in its Scan for Good initiative, which finds and remediates high-risk exposures in critical public infrastructure for free.

“In an early demonstration of its impact, the model uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide, identifying a severe risk that frontier models had missed,” Google said .

The announcement does not name the affected software or say whether the issue has been addressed.

On CWE-bench v1, a vulnerability remediation benchmark developed by Collinear AI, Argon tied for first place with a score of 68%, alongside OpenAI’s GPT-6 Astra and xAI’s Grok 4.7.

The company also claims improvements in vulnerability discovery compared to Gemini 3.8 Flash Cyber. On Google’s internal vulnerability benchmark, Argon found a wide range of exposures across complex codebases written in 20 programming languages.

On Wiz’s internal black-box penetration testing benchmark, which targets live web systems without access to source code, Argon outperformed Gemini 3.8 Flash Cyber. Google says it was better at discovering the attack surface, identifying vulnerabilities, and producing proof-of-concept evidence.

Google is strengthening safeguards before a wider release

“Safely releasing frontier capabilities at this level requires a phased approach,” Google said, noting that it is taking part in the US government’s voluntary process for pre-release model access.

For the broader rollout, Google says the model is designed to refuse requests that could enable cyber or chemical, biological, radiological, and nuclear (CBRN) attacks, while still supporting legitimate dual-use scientific research.

These safeguards include improved techniques for monitoring the model’s internal activations to spot misuse. Google also describes Argon as its most resilient model yet against indirect prompt injection.

“In order to prevent Argon from stepping out of bounds to try to accomplish a task in a way that goes beyond the user’s intentions, we are deploying misalignment mitigations that monitor Argon’s chain-of-thought and actions and stop execution when necessary,” the company explained.

The company is also isolating and sealing its sandboxed environments before high-risk training or evaluations begin.

Related : Google: AI Is Changing the Pace and Profile of Vulnerability Discovery

Related : Anthropic Flags AI Agent Liability Risks as OpenAI Faces Hacking Lawsuit

Related : Trump Says Top Tech Firms Have Signed Accord to ‘Self-Police’ AI Development

Eduard Kovacs (@EduardKovacs) is senior managing editor at SecurityWeek. He worked as a high school IT teacher before starting a career in journalism in 2011. Eduard holds a bachelor’s degree in industrial informatics and a master’s degree in computer techniques applied in electrical engineering.

More from Eduard Kovacs

High-Severity Vulnerabilities Patched in OpenSSL, WolfSSL

New Spectre v2 Variant Exposes Intel, AMD, Arm CPUs to Data Leaks

Pentagon Personnel Agency Data Breach Impacts 3 Million People

OpenAI Calls Off GPT-6.1 Astra Launch, Details Safety Cases for Frontier Training

Apple Patches Zero-Day Linked to ‘Extremely Sophisticated Attack’

Nvidia Unveils AI Agent Safety Platform With Hardware-Based Watchdog

Citrix Confirms 2 NetScaler Zero-Days After Admins Pulled the Plug

Microsoft SharePoint Flaw CVE-2026-65660 Now Exploited in Attacks

FTC is Investigating OpenAI and Anthropic Over Possible Risks to Consumers

Google: AI Is Changing the Pace and Profile of Vulnerability Discovery

WatchGuard Patches Critical Fireware OS Code Injection Vulnerability

Government, Finance Orgs Targeted in Weeks-Long NetScaler Zero-Day Attacks

Chrome, Firefox Updates Patch Over 100 Vulnerabilities

Anthropic Flags AI Agent Liability Risks as OpenAI Faces Hacking Lawsuit

Russian APT Star Blizzard Uses ‘RedFlick’ Infection Chain in Recent Attacks

ShinyHunters Defiant After FBI Calls on Members to Come Forward

Flipboard Whatsapp Whatsapp Email