AI Coding Tools Linked to 41% More Bugs and 2.74x Higher Security Flaws

Studies show that teams adopting AI coding assistants like Copilot, Cursor, and Claude Code have seen a 41% rise in bug rates, largely due to insufficient code review, according to research by Uplevel. CodeRabbit's findings further indicate that AI-assisted pull requests carry 1.7 times more issues than those written entirely by humans. Security vulnerabilities are approximately 2.74 times more prevalent in AI-generated code, often because models replicate common but insecure patterns found in public training data. The core problem is an imbalance: AI tools dramatically speed up code generation while review workloads remain unchanged, causing more defects to slip through. With roughly 95% of production code now containing AI-assisted content, the focus has shifted from whether to use these tools to how teams can maintain quality while retaining the speed benefits.