Google DeepMind
Build with Gemini
Try Gemini
Gemini 4 Argon without cyber guardrails

Frontier-level performance in vulnerability detection, and automated patching

Capabilities

Gemini 4 Argon provides a decisive advantage in today’s complex cybersecurity landscape.

The model is trained to be highly capable at cyber defense, has frontier-level performance in vulnerability detection, and automated patching.

Autonomous vulnerability discovery

Demonstrates frontier-level performance in autonomous vulnerability discovery. Capable of navigating complex codebases to uncover critical hidden vulnerabilities.

Penetration testing

Performs black-box penetration of web systems with access only to the running website and its public behavior, without application source code, to discover the attack surface, identify vulnerabilities, and produce validation evidence.

Automated patching

Generates validated, high-quality code fixes automatically after discovering vulnerabilities. So defenders can neutralize potential threats before attackers discover and exploit them.

Wiz is already using Argon for its “Scan for Good” initiative, which protects critical public infrastructure for free. Having had early access to the model, Wiz used Argon to protect critical healthcare infrastructure. The model went beyond just spotting a vulnerability - it proactively safeguarded hundreds of thousands of patient records.

Learn about Wiz
Learn about Scan for Good

Fairwind Program

The Fairwind Program gives trusted cyber defenders a critical head start against cyber threats, while making sure models are deployed responsibly.

Learn more

Performance

Gemini 4 Argon with cybersecurity defense capabilities, starting to roll out to a set of trusted defenders via the Fairwind Program, provides a decisive advantage in today’s complex cybersecurity landscape.

A bar chart displaying the CWE-bench v1 leaderboard scores. Grok 4.7, Gemini 4 Argon, and GPT-6 Astra are tied at the top with a score of 68.0%, followed closely by Claude Opus 5.5 at 67.0%, while other models score between 37.0% and 58.0%.A bar chart displaying the CWE-bench v1 leaderboard scores. Grok 4.7, Gemini 4 Argon, and GPT-6 Astra are tied at the top with a score of 68.0%, followed closely by Claude Opus 5.5 at 67.0%, while other models score between 37.0% and 58.0%.
A bar chart comparing real-world vulnerability discovery performance, showing Gemini 4 Argon with a score of 85.8% and Gemini 3.8 Flash Cyber with 71.0% on scanning source code across 20 languages.A bar chart comparing real-world vulnerability discovery performance, showing Gemini 4 Argon with a score of 85.8% and Gemini 3.8 Flash Cyber with 71.0% on scanning source code across 20 languages.
A bar chart comparing Wiz penetration test benchmark scores. Gemini 4 Argon achieves 70.9%, while Gemini 3.8 Flash Cyber scores 58.2%.A bar chart comparing Wiz penetration test benchmark scores. Gemini 4 Argon achieves 70.9%, while Gemini 3.8 Flash Cyber scores 58.2%.
This bar chart compares the "Gray Swan IPI" attack success rate across various AI models for k=1, 10, and 15 attempts, where a lower score is better. Gemini 4 Argon achieves the lowest attack success rate at 0.7%, followed closely by Claude Opus 5.5 and Claude Fable 5.1 at 1.0%, while Kimi K3 has the highest rate at 52.7%.This bar chart compares the "Gray Swan IPI" attack success rate across various AI models for k=1, 10, and 15 attempts, where a lower score is better. Gemini 4 Argon achieves the lowest attack success rate at 0.7%, followed closely by Claude Opus 5.5 and Claude Fable 5.1 at 1.0%, while Kimi K3 has the highest rate at 52.7%.