Google launches Gemini 4 Argon, starting with guardrail-free access for vetted cyber defenders
The new frontier model can find, validate and patch software flaws; Google says it already uncovered a critical vulnerability in hospital software used worldwide.
What launched: Google announced Gemini 4 Argon, its latest frontier AI model, and is initially rolling it out to trusted cyber defenders through its Fairwind Program, SecurityWeek and Engadget reported on Thursday. Google’s internal teams are also using the model. Broader access for developers, enterprises and Google AI Ultra subscribers is planned later, Engadget said.
Google says Argon is built for complex workflows in software engineering, finance, legal work, creative writing and cybersecurity defense. A spokesperson told Engadget the company sees Argon as comparable on key benchmarks to OpenAI’s GPT-6 Astra and Anthropic’s Opus. Independent firm Artificial Analysis said Argon matches Astra’s Intelligence Index score at about 60% of the cost per task at current discounted prices, with an introductory API price of $2 per million input tokens and $10 per million output tokens, Engadget reported. Argon’s hallucination rate was listed at 15%, versus 54% for Astra and GPT-6.1 Sol, according to Artificial Analysis.
Cyber defense focus: Google said it trained Argon to autonomously find, validate and patch critical software vulnerabilities, and that for trusted defenders and Google’s own teams it is releasing Argon “without cyber guardrails” so they can use its full defensive capabilities, SecurityWeek reported. Fairwind, launched in early September for governments, Google Cloud customers and cybersecurity partners, had more than 650 participants at launch.
In an early demonstration, Argon “uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide” that previous frontier models had missed, Google said, according to SecurityWeek. The announcement did not name the software or say whether the flaw has been fixed. On Collinear AI’s CWE-bench v1 vulnerability-remediation benchmark, Argon tied for first at 68% with GPT-6 Astra and xAI’s Grok 4.7.
Safeguards before a wider release: For the broader rollout, Google says Argon is designed to refuse requests that could enable cyber or chemical, biological, radiological and nuclear attacks, while still supporting legitimate dual-use research, SecurityWeek reported. The company said it is deploying mitigations that monitor the model’s chain of thought and actions and can stop execution when needed, and described Argon as its most resilient model yet against indirect prompt injection.
What this means: Google is shipping a frontier model first to a closed circle of cyber defenders — including without some of the usual restrictions — before a phased public release, betting that stronger defensive capability is worth the careful rollout.