How a hacker found a vulnerability with ChatGPT

The chatbot ChatGPT is currently on everyone’s lips. A hacker is now reporting that OpenAI’s artificial intelligence even helped him find a security hole.
ChatGPT was developed by the American company OpenAI and is based on the GPT-3 language model. If you ask the artificial intelligence questions, it gives direct answers — no selection of links, such as the Google search engine.
The field of application of ChatGPT is very large — and apparently can even help in the digital security sector.

ChatGPT was doubly helpful to hackers

French white hat hacker Lupine relied on ChatGPT when testing a hospital’s software system.
White hat hackers act as hackers for hire, constantly testing companies’ information infrastructure to ensure there are no security vulnerabilities. In doing so, they help companies to protect themselves from the harmful black hat hackers.
Lupine entered the program code of the software system to be examined at ChatGPT — and benefited from it. Because the chatbot not only explained the code to him, but also pointed out potential points of attack and thus helped to identify weak points in the system, as Gamestar reports. In a French YouTube video, Lupine explained his approach:

Facebook vulnerability also discovered

In the video, the hacker also explains how he fed Facebook program code to ChatGPT and the AI ​​found a security hole there that was discovered last year. ChatGPT even provided the program code that was able to break the vulnerability.
The white hat hacker Youssef Sammouda received 38,000 euros from the US company for discovering the Facebook vulnerability. Whether Lupine was also paid by the hospital for his discovery remained unanswered.

ChatGPT statements are sometimes “wrong and nonsensical”

ChatGPT will change the “type of work”, Nicolas Flores-Herr from the Fraunhofer Institute told WDR . He believes that AI is a similar far-reaching revolution as the Internet. However, he and other scientists also warn that ChatGPT does not always work properly.
“ChatGPT delivers statements that sound plausible. But sometimes, if you take a closer look, they are wrong and nonsensical,” warned Christian Schlereth, head of the Chair for Digital Marketing at WHU in Düsseldorf, in an interview with WDR.

© Von Christian Bernhard