Google has unveiled Gemini 4 Carbon, its latest AI model designed to identify software flaws, assess whether attackers could exploit them, and generate fixes This article explores security warnings ai. . The model, introduced on October 28, 2026, builds upon Gemini 4 Argon and focuses on complex security and coding tasks that can take days.
Google's internal teams have reported seeing it work through dead ends while tracing subtle bugs across large codebases and low-level systems, then developing patches that address their root cause.
Google reveals that Carbon achieved 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. These tests encompass three interconnected stages of security work: identifying flaws, assessing their impact, and fixing the underlying code. Google Launches Gemini 4 Carbon Carbon also leads the Artificial Analysis Intelligence Index with a score of 59, combining nine evaluations across coding, scientific knowledge, knowledge work, and general knowledge.
Google emphasizes that Carbon employs multiple safety layers, including misuse checks, prompt injection defenses, and monitoring of reasoning and actions. Previous reports on Gemini for Workspace prompt injection highlighted how hidden email instructions could lead to misleading security warnings in AI summaries. Separate coverage of Gemini data theft vulnerabilities revealed how poisoned content could expose saved information before Google addressed the reported issues.
Government testing is underway before a wider rollout.
