Recent data from "CodeRabbit" reveals that code produced using artificial intelligence tools contains more issues and security vulnerabilities compared to code written by humans.
Increase in issues in AI-generated code
Average issues in Pull Requests produced by artificial intelligence: 10.83 issues.
In human-written code: only 6.45 issues.
It was also found that AI-generated code contains:
- 1.4 times the critical issues.
- 1.7 times the major issues.
Most common types of issues
The AI-generated code led to issues in:
Security: weak password handling, unsafe references, XSS vulnerabilities, and unsafe deserialization.
Performance and logic: errors in logic correctness and performance errors.
Code quality and maintenance: issues in maintainability and organization.
Advantages despite the risks
- Artificial intelligence reduced spelling errors by 1.76 times less.
- Reduced code testing issues by 1.32 times less.
According to David Loker, director of AI at CodeRabbit: "Artificial intelligence tools increase productivity but also present predictable vulnerabilities that must be actively managed by organizations."
The study shows that AI-generated code is not a substitute for humans but requires thorough human review to ensure safety and quality.
At the same time, artificial intelligence can accelerate tedious processes and enhance productivity in the early stages of code development, while developers’ roles shift to managing and reviewing this code instead of writing it entirely.
With improvements in AI models like the GPT family, errors are expected to decrease over time, but human review will remain essential to ensure the quality and safety of software.
