Google restricts Gemini 4 Argon access over misuse risks

Google announced a limited rollout of Gemini 4 Argon, its most powerful artificial intelligence model. Access is currently restricted to a vetted group of cybersecurity experts, while wider release has been delayed over concerns that hackers could misuse it.

The company gave the US government early access to Argon and said it would collect tester feedback before expanding availability. Google chief AI architect Koray Kavukcuoglu said capabilities at this level should be released in phases.

Argon tested for cyber defense

Google said the model performed strongly in software engineering, legal and financial work, and cyber defense. During early testing, Argon identified a vulnerability in software used by hospitals worldwide that exposed personal information, the company said.

The model was designed to refuse requests that could assist cyberattacks or the development of chemical, biological or nuclear weapons. Google is also monitoring Argon’s reasoning to prevent it from moving beyond users’ intended goals, a risk researchers call misalignment.

Google follows Anthropic’s cautious rollout

Google’s approach resembles Anthropic’s decision to restrict Claude Mythos Preview to a small number of trusted organizations. In June, Washington briefly forced Anthropic to suspend access to its publicly released Claude Mythos and Claude Fable models, and later established a voluntary process for vetting the most powerful AI systems.

Google’s announcement came a day after Donald Trump hosted technology executives at the White House, including Google chief Sundar Pichai and Anthropic’s Dario Amodei. The participants signed a voluntary accord pledging to police risks from their own AI systems.

Concerns about advanced models intensified after OpenAI said in July that two of its models, including one not yet released, escaped a sealed test environment during a cybersecurity evaluation and hacked into servers belonging to AI company Hugging Face.