UK AI Security Institute tests Gemini 4 Argon and warns of breach risks

The United Kingdom AI Security Institute is conducting evaluations on more than 55 advanced artificial intelligence systems, including Google's unreleased Gemini 4 Argon model, according to shattered.io. UK AI minister Kanishka Narayan confirmed on October 12, 2026, that the government body is running these tests before the models are released to the general public. shattered.io reported that the minister framed the announcement around national cyber defence and warned that a cyber breach remains likely even as voluntary safety agreements hold. [1]

The evaluations aim to assess potential security risks associated with the newest wave of frontier artificial intelligence before deployment. shattered.io reported that government officials are publicly cautioning AI developers about the safety and security risks inherent in their unreleased systems. [1]

The AI Security Institute operates as a research organisation within the United Kingdom's Office for the Prime Minister and Cabinet. It was established as the first institution of its kind to equip governments with a scientific understanding of risks posed by advanced artificial intelligence through research, testing, and the development of mitigations. [4]

The government's blunt warnings highlight growing concern over national cyber security as frontier models increase in capability. shattered.io noted that the pre-release testing program reflects heightened official scrutiny over whether voluntary developer commitments are sufficient to prevent severe security incidents. [1]