Google has announced a cautious rollout of its latest artificial intelligence model, Gemini 4 Argon, restricting public access to a selected group of cybersecurity experts due to concerns about potential misuse by hackers. The company said the phased release will allow specialists to evaluate the model’s capabilities and identify possible security risks before it becomes widely available.
Google’s chief AI architect, Koray Kavukcuoglu, said that introducing advanced AI capabilities safely requires a gradual approach. The company is also providing the US government with early access to the model and plans to collect feedback from authorised testers before considering a broader release.
The move follows a similar strategy adopted by rival AI company Anthropic, which has restricted access to its advanced Claude Mythos Preview model to a limited number of trusted organisations. US authorities have also introduced a voluntary review process for evaluating powerful AI systems before their public release.
The announcement comes amid growing concerns that advanced AI technology could be exploited to launch cyberattacks against banks, hospitals and government institutions. Google says Gemini 4 Argon demonstrates strong capabilities in software engineering, legal and financial tasks, and cybersecurity, particularly in identifying and fixing critical software vulnerabilities.
According to Google, early testing has already helped uncover a software flaw affecting hospitals worldwide that could expose sensitive personal information. The company says the model includes safeguards designed to reject requests involving cyberattacks and the development of chemical, biological or nuclear weapons.
Google is also monitoring the model’s reasoning to reduce the risk of misalignment, a situation in which an AI system acts beyond its intended objectives. Similar safety measures have been introduced by other leading AI developers, including Anthropic and OpenAI.
Concerns over AI security have intensified following reports that OpenAI models breached a controlled testing environment during a cybersecurity evaluation and accessed servers belonging to AI company Hugging Face. These developments have increased pressure on technology companies to strengthen safeguards before releasing increasingly powerful AI systems to the public.
