- “AI is fixing typographic errors but creating the combas of time,” says Apiiro
- 10 times more security findings stood out in June 2025 against December 2024
- AI is 4 times faster than humans, and can also reduce syntax errors and logical errors
New research has affirmed that despite its promised advances to help code faster and more effectively, the use of AI tools in coding may be throwing security problems in all areas.
Apiiro has quantified the vulnerabilities to which companies could be exposed by using the code generated by AI, finding that developers assisted by AI-AI can write 3-4x more code than their peers, however, their code introduced 10 times more security problems.
Vulnerabilities varied in severity, and although they were not always exploitable errors, the quality of the code still saw a remarkable fall.
The code generated by AI can really be of poor quality
Apiiro found insecure patterns, exposed secrets, new dependencies and erroneous configurations of the cloud in the code generated by AI, as well as the exploitable errors that represent the greatest risk.
By June 2025, the code generated by AI was throwing 10,000 new security findings per month, an increase of ten times in the six months from December 2024.
However, artificial intelligence also has many clear benefits. According to the data, the syntax errors in the code written by the fell by 76% and the logical errors fell into more than 60%. On the contrary, privileged climbing routes increased by 322% and architectural design defects increased 153%, problems that reviewers fight to detect.
“The AI is fixing typographic errors but creating the timas,” the researchers summarized.
Apiiro also pointed out that the developers assisted by AI exposed sensitive keys almost double that their unavailable partners. “Because attendees generate major changes in multiple files, a single credential can be spread in multiple services or configurations before someone realizes,” he explains.
All this in an era in which AI coding is actually being mandatory, not only supported, by companies such as Coinbase and Citi. The great technological leaders also indicate that about a third or more of their new code is generated by AI.
Companies should consider implementing more safeguards every time they use code generated by AI, since it also serves as a reminder that human supervision, logic and experience cannot be overlooked.
