Google has announced Gemini 4 Argon, a new AI model it says performs strongly on complex software engineering, business and cybersecurity tasks. Access is initially limited to a group of “trusted cyber defenders” while Google strengthens safeguards, making the announcement notable for its cautious, staged release.

Google says Argon is already being used internally, including for large codebase migrations, and claims its benchmark results compare favourably with models from OpenAI and Anthropic. It plans safeguards against misuse and prompt injection, and monitoring for misalignment before wider access; the launch follows OpenAI’s recent DevDay announcements and comes after DeepMind chief Koray Kavukcuoglu took up his role in August.