From Ars Technica · · 1m 30s
Google's new AI model is too risky to release
Gemini 4 Carbon beats rivals on benchmarks but Google won't let anyone use it yet, citing safety concerns.
In brief
Gemini 4 Carbon beats rivals on benchmarks but Google won't let anyone use it yet, citing safety concerns. Google announced Gemini 4 Carbon, a frontier AI model it says outperforms competitors in coding and cybersecurity, but is withholding release pending government testing due to safety risks. Originally reported by…
Google releases Gemini 4 Carbon
Just two weeks after rolling out Gemini 4 Argon, Google announced Gemini 4 Carbon. This is an improvement on Argon, not a Flash equivalent, and Google is working on a staged release.
Carbon excels at coding and security
Gemini 4 Carbon reportedly makes big gains in cybersecurity applications.Google says the model shows a significant leap in vulnerability detection, validation, and remediation.Carbon is twice as likely to generate code free of vulnerabilities versus Argon.
“"significant leap" in vulnerability detection, validation, and remediation”
Carbon tops benchmark leaderboards
Carbon beats GPT-6 Astra, Claude Opus five point five, and Claude Fable five point one in almost every test.
Carbon resists all prompt injection attacks
In the Grey Swan evaluation, Carbon resisted every attempted prompt injection attack, scoring zero percent.
Google delays public release for safety
Google is taking a deliberative approach to Carbon's release, working within the voluntary testing framework established by the US government. The model will roll out through its Fairwinds cyber defense program once testing is done.
“"deliberative approach"”
Early access pricing and timeline unclear
Carbon will probably arrive first for AI Ultra subscribers and developers using the Gemini API.
The takeaway
Google announced Gemini 4 Carbon, a frontier AI model it says outperforms competitors in coding and cybersecurity, but is withholding release pending government testing due to safety risks.
