Menu Close

Google unveils Gemini 4 Argon with phased cyber-partner rollout amid U.S. safety evals

Illustration on a chalk-white background. Across the center and lower band, a sealed midnight-plum frontier capsule sits behind a glacier-cyan pass-gate arch. A narrow defender-first lane strip runs left to right under the gate toward a small cohort of blank unbranded partner chips, while a soft-amber pre-release eval hold bar blocks a wider public-access path on the right. Quiet lower chips read Fairwind cohort, cyber first, safety eval, and phased roll out.

Google on Wednesday unveiled Gemini 4 Argon, its new frontier model, and said a phased rollout will begin with trusted cybersecurity partners while the company works with the U.S. government on pre-release safety evaluations.

In a company post, Google DeepMind SVP and Chief AI Architect Koray Kavukcuoglu said Argon is rolling out to trusted cyber defenders through Google’s Fairwind Program and that the company is engaged in the U.S. government’s voluntary process for pre-release model access as it gradually expands availability. CNBC reported that Alphabet launched the model to select cyber partners and plans to scale safeguards, including defenses against misuse and prompt injection, before a public launch. “Starting this rollout in this way gives us more confidence, but also enables us to put a model that is trained and strong in cyber defense in the hands of defenders as soon as possible,” Tulsee Doshi, Google’s Gemini model product lead, told CNBC.

Google said Argon sets a new state of the art on DeepSWE v1.1 for real-world long-horizon software engineering, ties for first on CWE-bench v1 for vulnerability remediation, and leads the Vals Index measuring economic impact across finance, coding, legal, and tax work. CNBC reported that, on cybersecurity evaluation benchmarks, Argon ties OpenAI’s GPT-6 Astra and Grok 4.7, and that it leads GPT-6 Astra and Anthropic’s Fable 5.1 on the Vals Index. Those peer rankings are Google and CNBC framing, not independent AI Tech Daily scores.

Google said Argon is already used internally, including memory optimizations across its data centers and work with quantum computing researchers, and that output context rises to 1 million tokens. For approved Fairwind defenders and internal teams, Google said it will release Argon without cyber guardrails so they can use its full defensive cybersecurity capabilities. Broader access is slated to start with paid API customers and Google AI Ultra subscribers after early testing and guardrail iteration.

AI Tech Daily reported on Sep. 24 that Kavukcuoglu said Gemini 4 was in post-training with an early-release aim. Wednesday’s Argon unveil is that frontier ship under a cyber-first, safety-gated plan.

Sources

0 0 votes
Article Rating
Subscribe
Notify of
0 Comments
Inline Feedbacks
View all comments
0
Would love your thoughts, please comment.x
()
x