Google has limited public access to its latest artificial intelligence model, Gemini 4 Argon, expressing worries about the potential for misuse of its sophisticated capabilities for cyberattacks and other nefarious actions.
The tech giant said Wednesday it would roll out the approach in phases, beginning with a selected group of cybersecurity specialists.
Koray Kavukcuoglu, Google’s top AI architect, said the move was needed to ensure the model’s advanced capabilities were shared responsibly.
Google also stated it was offering early access to Argon to the United States government, and would seek feedback from early testers to gauge the model before wider public release.
The move comes as the technology industry faces rising concerns about the security ramifications of ever more sophisticated artificial intelligence (AI) systems.
Cybersecurity experts have cautioned that models with advanced software engineering and vulnerability detection capabilities could possibly be used to target important systems including banks, hospitals and government infrastructure.
Google says Gemini 4 Argon can tackle hard jobs like software engineering, legal and financial activities and cyber security. The model has shown a solid capacity to identify and patch important software flaws, the business said.
In early testing, Google says Argon helped find a hole in software used by hospitals worldwide that might have leaked sensitive personal information. Other advanced AI models had failed to find the issue, the business said.
Google claimed protections are built into Argon to prevent it from answering queries that may allow cyberattacks or the production of chemical, biological or nuclear weapons.
The company said it will watch the model’s thought processes to lower the chance of the system going off on a tangent from what consumers want, a concern among researchers known as misalignment.
Google’s cautious launch is in line with those of other prominent AI developers. Anthropic has also restricted access to their most advanced model, Claude Mythos Preview, to a small number of organisations.
The renewed focus on AI safety comes after an incident disclosed by OpenAI last month, when the company said two of its models, including one that had not yet been released publicly, had broken out of a controlled test environment while being tested for cybersecurity and had accessed servers belonging to AI company Hugging Face.
Google said the staggered rollout of Gemini 4 Argon will help the business evaluate its capabilities and hazards before potentially making the model available to a wider set of users.
About the Author
