AI code errors are more serious than those made by human programmers, study finds

The report notes that AI-generated pull requests (PRs) have an average of 10.83 issues, compared to 6.45 in “human” projects. The use of AI increases the number of errors by almost 1.7 times, which significantly delays the code review process and creates additional risks to software quality, The Register writes .

Issues caused by AI-generated pull requests (PRs) are typically more serious than errors made by humans. According to the report, AI-generated PRs contain, on average, 1.4 times more critical issues and 1.7 times more serious errors than those written by humans.

Therefore, machine-generated code requires reviewers to address a large number of issues, which are also more serious than those found in human-written code.

The report also says that AI-generated code outperforms human-written code across all major problem categories:

• logic and correctness — 1.75 times more;
• quality errors and code support — 1.64 times more;
• security flaws — 1.57 times;
• productivity problems — 1.42 times.

In addition, this also applies to specific security issues:

• 1.88 times more likely to contain improper password handling;
• 1.91 times more likely to have dangerous links to objects;
• added XSS vulnerabilities 2.74 times more often;
• implemented unsafe deserialization 1.82 times more often.

At the same time, AI showed better results in spelling and testing:

• spelling errors occurred 1.76 times more often in PRs created by humans;
• also, human-generated code had 1.32 times more testing problems.

“These results confirm what many engineering teams have been experiencing throughout 2025,” David Locker, Director of AI at CodeRabbit, said in a statement. “AI tools for coding are dramatically increasing productivity, but they also introduce predictable and measurable weaknesses that companies must actively address.”

“Beginners learn to assemble something ‘on their knees’ and consider it the norm.” An experienced engineer explained the key dangers of vibecoding and why experience, structure, and code purity are already becoming currency

On the topic

“Beginners learn to assemble something ‘on their knees’ and consider it the norm.” An experienced engineer explained the key dangers of vibecoding and why experience, structure, and code purity are already becoming currency

Swedish vibe-coding startup Lovable reports significant user growth — three times more than in the summer

On the topic

Swedish vibe-coding startup Lovable reports significant user growth — three times more than in the summer

32% of senior developers say half of their code is written with AI — twice as many as juniors

On the topic

32% of senior developers say half of their code is written with AI — twice as many as juniors

Stack Overflow study shows developers are frustrated with AI coding tools, and here's why

On the topic

Stack Overflow study shows developers are frustrated with AI coding tools, and here's why

Read the country's main IT news in our Telegram

On the topic

Read the country's main IT news in our Telegram