- “The AI repairs typing faults but creates time loops,” says Apiiro
- 10x other security results were highlighted in June 2025 compared to December 2024
- AI is 4x faster than humans and can also reduce syntax errors and logical bugs
New research has said that despite its promised progress to help code more quickly and more efficiently, the use of AI tools in coding could launch security problems at all levels.
APIIRO has quantified the vulnerabilities to which companies could expose themselves using code generated by AI, the search for developers assisted by AI was able to write 3 to 4x more than their peers, but their code has introduced 10x more security problems.
Vulnerabilities varied in severity, and although they were not always exploitable bugs, the quality of the code has always experienced a significant drop.
The code generated by AI can be really poor quality
APIIRO has found unsecured models, exposed secrets, new dependencies and cloud confusion errors through the code generated by AI, as well as exploitable bugs that represent the greatest risk.
In June 2025, the code generated by the AI launched 10,000 new safety conclusions per month, an increase of ten times in the six months from December 2024.
However, artificial intelligence also has many clear advantages. According to data, syntax errors in the code written by AI fell 76% and logical bugs have dropped by more than 60%. Conversely, the climbing routes of privileges increased by 322% and architectural design defects increased by 153% – the problems that examiners have difficulty identifying.
“The AI is repairing typing faults but creates the time loops,” summed up the researchers.
APIIRO also noted that the AI -assisted developers exhibited sensitive keys almost twice more often than their unaccompanied peers. “Because assistants generate large multi-fichiers changes, a single diploma can be propagated on several services or configurations before anyone noticed,” explains the.
All this at a time when the coding of AI is really mandated – not only supported – by companies like Coinbase and Citi. The big technology leaders also indicate that around one third or more of their new code is generated by AI.
Companies should consider implementing new guarantees each time they use code generated by AI, as it also recalls that human surveillance, logic and experience cannot be overlooked.
