Google says Gemini 4 Carbon beats its predecessors at finding and fixing security flaws

Google said its frontier AI model Gemini 4 Carbon outperforms its predecessors at finding and fixing security flaws, with safeguards built on those of its predecessor, Argon. Google reported that its teams have seen Carbon spend days investigating and fixing bugs across large codebases and low-level systems, and that the model was trained to handle complex tasks and high-pressure situations. The claims were evaluated on SEC-bench Pro, according to Google.