Google limits Gemini 4 Argon rollout over safety concerns

Google is keeping Gemini 4 Argon away from public release for now, limiting access to a vetted group of cybersecurity experts while it assesses safety risks. Koray Kavukcuoglu, Google’s chief AI architect, said releasing frontier capabilities requires a phased approach. Google is also giving the US government early access and plans to gather feedback before a wider rollout.

The company said Argon performs strongly in software engineering, legal and financial work, and cyber-defense, including finding and fixing critical software flaws. Early testers used it to uncover a vulnerability in hospital software used worldwide that exposed sensitive personal information, a flaw Google said other advanced models had missed.

The restricted launch reflects wider concern that powerful models could be misused to attack banks, hospitals and government systems. Google said Argon is built to refuse requests that could support cyber-attacks or help develop chemical, biological or nuclear weapons, and that it is monitoring the model’s reasoning to prevent misalignment. Anthropic has taken a similar limited-release approach, while OpenAI and Anthropic have added comparable safeguards to their advanced models.