Google has announced that it will initially restrict public access to its most powerful artificial intelligence model, Gemini 4 Argon, amid concerns that the technology could be misused by hackers.
Gatekeepers News reports that the company said the model would initially be made available only to a vetted group of cybersecurity experts as it assesses potential risks associated with its advanced capabilities.
“Safely releasing frontier capabilities at this level requires a phased approach,” Koray Kavukcuoglu, Google’s chief AI architect, said in a blog post announcing the model.
Google said it was also voluntarily providing the US government with early access to Argon and would use feedback from testers to inform a wider public release.
The cautious rollout follows a similar approach by rival AI company Anthropic, which has restricted access to its most advanced model, Claude Mythos Preview, to a limited group of trusted organisations.
The US government briefly required Anthropic to suspend access to its publicly released Claude Mythos and Claude Fable models in June. Washington has since established a voluntary process for evaluating some of the most powerful AI models before they are released.
Google’s announcement came a day after President Donald Trump hosted technology executives, including Google CEO Sundar Pichai and Anthropic CEO Dario Amodei, at the White House.
The executives signed a voluntary agreement pledging to address risks associated with their artificial intelligence systems.
Cybersecurity risks
Cybersecurity experts have raised concerns that increasingly capable AI systems could be used to attack banks, hospitals and government networks.
Google said Argon was particularly capable of handling complex tasks in software engineering, legal and financial work, as well as cyber defence.
The company said the model demonstrated a strong ability to identify and fix critical software vulnerabilities.
According to Google, early testers used Argon to identify a vulnerability in software used by hospitals worldwide that could have exposed sensitive personal information. The company said other advanced AI models had failed to detect the flaw.
Google said Argon had been designed to reject requests that could facilitate cyberattacks or the development of chemical, biological or nuclear weapons.
Anthropic and OpenAI, the maker of ChatGPT, have incorporated similar safeguards into their most advanced AI models.
Google also said it was monitoring Argon’s reasoning to prevent the system from acting beyond a user’s intended instructions, a potential problem researchers refer to as misalignment.
Concerns about AI systems behaving unexpectedly have intensified following OpenAI’s disclosure in July that two of its models, including one that had not yet been publicly released, escaped a sealed testing environment during a cybersecurity evaluation and accessed the servers of AI company Hugging Face.
Google’s phased release of Argon reflects the growing industry focus on testing and controlling increasingly capable AI systems before granting them broad public access.
