- 96% of developers don’t fully trust AI, 52% don’t always check for errors
- The majority of ChatGPT and Perplexity users access AI through a personal account
- Conversely to trends, data exposure and vulnerabilities are big concerns
Sonar’s latest State of Code Developer Survey has found almost all (96%) developers say they don’t fully trust AI-generated code to be functionally correct, despite widespread use, and even worse, many don’t check their AI-generated code properly.
Currently, around 42% of developers’ code is said to be AI-generated (up a significant amount from just 6% in 2023), but this is expected to rise to around 65% by 2027.
And yet not even half (48%) of them always check AI’s work before committing, highlighting a huge potential for bugs and vulnerabilities to be introduced.
Developers aren’t checking AI-generated code before using it
While three in five (59%) say that they give ‘moderate’ or ‘substantial’ effort into checking AI-generated code, two in five (38%) agree its takes more time to verify than the equivalent human-written code. And, because the code generated uses a whole internet load of data to inform its output, three in five (61%) agree that it often looks correct, but isn’t.
This study corroborates with a separate piece of recent research published by CodeRabbit, revealing that AI generates 1.7x more issues (and 1.7x more major issues) than humans.
Current trends show AI tools being most used across prototyping (88%) and internet production software (83%), which might not appear to be so critical, but nearly as many use it for customer-facing apps (73%). GitHub Copilot (75%) and ChatGPT (74%) are by far the most commonly used assistants.
But taking this information one step further, Sonar found more than one in three (35%) developers use their own personal accounts rather than work-approved ones – a figure that rises to 52% among ChatGPT users, and 63% for Perplexity fans. This presents yet another risk to potentially confidential or sensitive company information.
In fact, despite what Sonar found to be true about AI usage among developers, data exposure (57%), small vulnerabilities (47%) and severe vulnerabilities (44%) are among the highest concerns.
“Generating code faster is only half the battle,” the report concludes. “The real value comes from being able to trust and verify that code efficiently.”
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
