Humans are better coders than AI, CodeRabbit concludes
Generating code using AI tools can accelerate your work, but it also comes with increased risks. Research indicates an increased number of issues that require review when an AI code-generation tool is used. At the same time, these tools are likely to introduce more severe issues.
Image by Cybernews
Generating code using AI tools can accelerate your work, but it also comes with increased risks. Research indicates an increased number of issues that require review when an AI code-generation tool is used. At the same time, these tools are likely to introduce more severe issues.
That’s what CodeRabbit says in the latest edition of its State of AI vs. Human Code Generation Report.
Security researchers analyzed 470 open-source GitHub pull requests, including 320 AI-co-authored PRs and 150 human-only PRs. In short, they found that AI tools accelerate code output, but also amplify the number of (critical) mistakes.
For those who don’t know, a pull request (PR) is a way for developers to improve the quality and safety of their code. A developer can ask others to review their codebase, or other developers can propose changes.
According to CodeRabbit’s researchers, AI-generated PRs contained 1.7x more issues overall. On average, AI-generated PRs include about 10.83 issues per PR, compared with 6.45 issues in human-generated PRs.
At the same time, high-issue outliers were much more common in AI PRs, creating heavy review workloads.
Researchers also discovered that AI tools produce 1.4x more critical issues compared to humans, and 1.7x more major issues. These include logic mistakes, flawed control flow, and misconfigurations.
Furthermore, researchers found that security issues were higher with AI code-generating tools compared to humans. AI-generated code was more likely to introduce improper password handling (1.88x), more likely to make insecure object references (1.91x), and more likely to add cross-site scripting or XSS vulnerabilities (2.74x).
“AI coding tools are powerful accelerators, but acceleration without guardrails increases risk. Our analysis shows that AI-generated code is consistently more variable, more error-prone, and more likely to introduce high-severity issues without the right protections in place,” CodeRabbit says in a blog post.
“These findings reinforce what many engineering teams have sensed throughout 2025. AI coding tools dramatically increase output, but they also introduce predictable, measurable weaknesses that organizations must actively mitigate,” David Loker, Director of AI at CodeRabbit, said in a statement.
Unlock more exclusive Cybernews content on YouTube.