AI-generated code contains 1.7 times more security vulnerabilities than code written by humans, according to a study published by Nithin Kunimmal's team at @coderabbitai. The finding runs counter to a common assumption that automated coding tools should produce fewer errors than human developers, who get tired, forget syntax, or lose track of dependencies. The issue is not about memory or fatigue. It comes down to context. AI coding agents write code without full knowledge of how an organization works. That context is rarely in one place. It lives in Slack threads, email chains, internal documentation, and the institutional knowledge of individual team members. Today's AI tools cannot access or synthesize all of it. Nithin says this gap is the core driver of vulnerabilities. Until AI can reliably absorb the full context of a team's codebase, its conventions, and its history, the output will carry risks that even a distracted human developer might have avoided.