Alphabet Unveils Gemini 4 Argon for Cybersecurity and Enterprise

Alphabet Inc. unveiled Gemini 4 Argon on Wednesday, positioning the model as its most advanced artificial intelligence tool for high-stakes professional work in coding, cybersecurity, finance, and law. The model achieved a tie with OpenAI's GPT-6 Astra and Grok 4.7 for the highest score on the CWE-bench for patching security vulnerabilities and outperformed GPT-6 Astra and Anthropic's Fable 5.1 on the Vals Index.

Google engineers have already applied Argon internally to free over 300 tebibytes of data center memory without requiring new hardware. To prevent misalignment, the company implemented systems to track the model's chain of thought and block attempts to evade monitoring.

The rollout is proceeding in phases through the Fairwind Program for vetted governments and cyber authorities. Google is collaborating with the U.S. government on pre-release safety evaluations, a process following a voluntary AI safety agreement signed by CEO Sundar Pichai and President Donald Trump at the White House. Tulsee Doshi, the Gemini model product lead, stated that providing a model of this caliber to defenders is meaningfully important and that the phased rollout enables the company to deploy a strong cyber defense tool as soon as possible.