Claude-generated rsync patches introduced more bugs than human code

A detailed analysis found that Claude-written contributions to rsync contained higher defect rates than patches from experienced maintainers, raising questions about relying on LLMs for critical infrastructure maintenance. The research suggests AI assistance in low-level systems code requires stronger review practices, not just acceptance.