Google restricts initial access to Gemini 4 Argon as it builds safeguards
The model is aimed at software engineering, enterprise work and cybersecurity. Google says access will expand gradually after further safety measures.
Google has announced Gemini 4 Argon, a frontier AI model that will initially be available only to selected “trusted cyber defenders” as the company strengthens its safeguards.
Google DeepMind chief AI architect Koray Kavukcuoglu said the model is designed for complex software engineering, enterprise work, including legal and finance, and cybersecurity defence. The company has not specified how long the limited-access phase will last or when broader access may begin.
Kavukcuoglu said Argon is already used in Google’s internal workflows, including large-scale codebase migrations. Google also presented benchmark comparisons with competing models, but the supplied material does not include an independent assessment of those results.
Before a broader rollout, Google plans measures to guard against misuse and prompt injection attacks, alongside monitoring for misalignment. Kavukcuoglu said Google is engaging in the US government’s voluntary pre-release model access process as access expands gradually.
