GOOGLE RELEASES GEMINI 4 CARBON TO SELECT CYBERSECURITY COMPANIES

Google has released Gemini 4 Carbon to a select group of cybersecurity defence companies, with wider access planned through its Fairwind Program. The AI model follows Gemini 4 Argon, which Google released earlier this month. Google says Carbon makes a “significant leap” in finding software vulnerabilities, checking them and helping to fix them. The company also claims gains in coding and tasks that require an AI system to complete several steps. Google has not announced general availability.

Google says Carbon leads the Artificial Analysis Intelligence Index (v5.0), ahead of Claude Opus 5.5 and GPT-6 Astra. It also places Carbon first on SEC-bench Pro, which tests the discovery of real-world bugs, and ExploitGym, which tests the use of vulnerabilities to produce exploits. According to Google, Carbon is twice as likely as Argon to generate code free of vulnerabilities. Google’s Chrome team says Carbon produces “high-quality” patches at 10 times the rate of models made outside Google. The company says Carbon gained 89 points on WebDev Arena and now ranks fourth there.

Google says its teams have seen Carbon work for days on complex engineering tasks, continuing after unsuccessful approaches. It also claims the model is better at avoiding shortcuts and collaborating with users. In a Grey Swan evaluation cited by Google, Carbon resisted every attempted prompt-injection attack; Argon, Claude Opus 5.5 and GPT-6 Astra recorded attack success rates of 0.7%, 1% and 8.5% respectively. Those results reflect the tests and measures Google cited, rather than performance across all uses. Google says it will extend access to other Fairwind Program partners, but has given no timetable.