Google unveils Gemini 4 Argon AI for cybersecurity and complex workflows

Google has unveiled Gemini 4 Argon, a new frontier artificial intelligence model designed for complex reasoning, long-horizon software engineering, and advanced cyber defense. A key technical feature of the model is its expanded generation boundary, which increases the output token limit from 64,000 to 1 million tokens, allowing the system to analyze, plan, and self-correct across massive projects in a single turn.

In cybersecurity applications, Gemini 4 Argon is capable of autonomously locating, validating, and repairing vulnerabilities across more than twenty programming languages. The model achieved a 68% score on the CWE-bench v1 benchmark, which evaluates the ability to identify and fix critical security flaws. Google is deploying the model through its Fairwind program, which provides participants with specialized access to its defensive capabilities. The security firm Wiz has already utilized the model to discover a critical vulnerability in hospital software that had not been detected by other models.

The announcement follows the signing of the White House Agreement on Superintelligence by Google CEO Sundar Pichai and U.S. President Donald Trump, a voluntary commitment to implement internal audits and controls for critical areas such as cybersecurity and biological threats.