Google Launches Gemini 4 Carbon, an AI Model Built for Cybersecurity

Google launched Gemini 4 Carbon on October 28, 2026 — a new AI model purpose-built for cybersecurity that continued to lead AI coverage into October 31. The model is designed to find software flaws across large codebases, test whether those flaws can be exploited by real attackers, and then produce code patches that address the root cause — work that typically occupies security teams for days. On benchmarks, Carbon scored 88.2% on SEC-Bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation testing, and 71.0% on CWE-Bench for code patching. It more than doubled its predecessor's success rate at generating secure code and achieved a 0.0% attack-success rate on indirect prompt-injection tests, while reducing high-severity misalignment alerts by roughly 80%. The model includes multiple safety layers: misuse checks, prompt-injection defences, and continuous monitoring. Introductory pricing is $2 per million input tokens and $10 per million output tokens (with a 95% discount on cached inputs); standard rates will later rise to $4 and $20 respectively. Initial access is restricted to selected cyber defenders, with broader rollout planned through Google's Fairwind Program before eventually reaching developers and general users; a public release timeline was not announced.