Google has restricted public access to its most powerful artificial intelligence model, Gemini 4 Argon, citing concerns that hackers could exploit its advanced capabilities in cybersecurity.
On Wednesday, Google announced that it will not be releasing its most powerful AI model to the general public for the time being. Instead, Gemini 4 Argon will only be available to a vetted group of cybersecurity experts to prevent its misuse by hackers.
Korey Kavukchoglu, Google's lead AI architect, wrote in the blog announcing the model that 'a safe rollout of such advanced capabilities requires a phased approach.'
Google also stated that it is voluntarily providing early access to the model to the US government and gathering feedback from testers before making it widely available.
This cautious launch process is similar to the approach taken by competitor Anthropic, which has kept its most advanced model, Claude Mythos Preview, limited to a small number of trusted organizations.
Previously, Washington was forced to temporarily suspend access to Anthropic's publicly released models Claude Mythos and Claude Fable in June, after which a voluntary vetting procedure for the most powerful AI models was established.
The announcement followed just one day after President Donald Trump met with leading technology executives, including Google's chief representative Sundar Pichai and Dario Amodei from Anthropic, at the White House. At this meeting, they signed a voluntary agreement committing to controlling the risks associated with their respective AI systems.
Cybersecurity experts express concern that this advanced technology could be used to hack banks, hospitals, and government systems.
Google emphasized that Argon excels at complex tasks in software development, legal and financial work, as well as cyber defense, possessing high potential for detecting and eliminating critical software bugs.
Early testers used Argon to identify a vulnerability in hospital software worldwide that exposed confidential personal information—something other advanced models missed, according to Google.
Google also stated that Argon is designed to refuse requests that could facilitate cyberattacks or the development of chemical, biological, or nuclear weapons.
Similar safeguards have been implemented in Anthropic's most advanced models and OpenAI's ChatGPT maker.
Google mentioned that it monitors the model's logic to prevent it from deviating from the user's intended course, a problem researchers call misalignment.
This issue gained new urgency after OpenAI reported in July that two of its models, including an unreleased one, escaped a sealed testing environment during cybersecurity assessment and hacked Hugging Face servers.