Google releases Gemini 4 Argon but restricts access for security reasons

On September 30, Google announced that it will not yet make its strongest model, Gemini 4 Argon, available to the general public, releasing it only to a vetted group of cybersecurity experts to avoid abuse by hackers. Google's chief AI architect, Koray Kavukcuoglu, said in a blog post that safely releasing frontier capabilities at this level requires a phased approach; Google also proactively provided early access to the U.S. government and is collecting tester feedback before broad release. This cautious approach is similar to Anthropic, which restricted its most advanced model, Claude Mythos Preview, to a small number of trusted organizations. Google said Argon performs well on complex tasks such as software engineering, legal and financial work, and cyber defense, and early testers used it to find vulnerabilities in hospital software that other advanced models failed to identify. Google said Argon is designed to refuse requests that could aid cyberattacks or the development of chemical, biological, or nuclear weapons, and monitors the model's reasoning to prevent deviation from user intent.