Google has restricted public access to its latest artificial intelligence model, Gemini 4 Argon, citing concerns that its advanced capabilities could be misused to carry out cyberattacks.
The company unveiled the model on September 30 but is initially making it available only to a vetted group of cybersecurity experts and trusted partners through its Fairwind programme. Google has not announced a date for wider public access.
“Safely releasing frontier capabilities at this level requires a phased approach,” Google chief AI architect Koray Kavukcuoglu said in a company blog post announcing the model.
Google is also providing the US government with early access to Argon as part of a voluntary process for evaluating advanced AI models before wider release.
Cybersecurity capabilities drive safety concerns
Argon is designed to handle complex, extended tasks across areas including software engineering, legal and financial work and cybersecurity. Google says the model has particularly strong capabilities in identifying and fixing software vulnerabilities.
Early testers used Argon to identify a flaw in software used by hospitals that exposed sensitive personal information, according to Google. The company has also highlighted the model's ability to help cybersecurity defenders find and remediate vulnerabilities.
However, those same capabilities raise concerns about potential misuse. Advanced AI systems capable of identifying vulnerabilities could potentially be used by malicious actors to attack banks, hospitals, government systems and other critical infrastructure.
Google said Argon has safeguards designed to refuse requests that could facilitate cyberattacks or the development of chemical, biological or nuclear weapons. The company is also monitoring the model's reasoning to identify behaviour that could move beyond a user's intended instructions.
The controlled rollout comes as AI companies face growing pressure to assess the risks of increasingly capable models before making them broadly available.
AI firms adopt more cautious release strategies
Google's approach follows a similar move by other frontier AI developers to limit access to their most capable systems while safety testing continues.
Google launched its Fairwind programme in September to give governments and trusted cybersecurity partners controlled access to advanced AI capabilities for defensive purposes. The programme is intended to help organisations identify and address vulnerabilities in critical infrastructure and other systems.
The latest release also follows discussions between major AI companies and the US government over voluntary safeguards for powerful AI systems. Google CEO Sundar Pichai and other technology executives met with US President Donald Trump at the White House on September 29, a day before the Argon announcement.
The shift towards restricted access reflects a growing tension in the development of frontier AI: the same capabilities that can help organisations identify security weaknesses can also increase the potential impact of malicious use.
Google's decision means Argon will initially be tested in a controlled environment before the company considers broader access, allowing it to gather feedback from cybersecurity experts and other trusted users while continuing to strengthen its safeguards.