The ChatGPT crawler vulnerability in the ChatGPT API enables DDoS attacks on websites. German security researcher Benjamin Flesch recently discovered this, which has now been published on GitHub.
The research posted on GitHub shows that the ChatGPT API contains a vulnerability in the handling of HTTP POST requests to the backend API /attributions endpoint
. This endpoint allows an array of hyperlinks to be delivered via the URLs parameter.
Problem with hyperlinks
The vulnerability is caused by the lack of restrictions on the number of hyperlinks in a request. This means that cybercriminals can overload requests with URLs via the API. Moreover, OpenAI does not verify whether the hyperlinks point to the same source or are duplicates.
Security researchers say this vulnerability could allow hackers to launch DDoS attacks on arbitrary websites. By placing thousands of hyperlinks in a single request, they can abuse OpenAI’s servers to send a huge volume of HTTP requests to a specific Web site.
The massive influx of concurrent connections can also cause the underlying infrastructure of the attacked websites to be overloaded and suffer damage. In these ways, DDoS attacks are enabled.
Strong criticism of OpenAI
The vulnerability is extra dangerous because the mechanisms of the OpenAI API do not restrict duplicate requests and do not check for them. This makes the API a powerful attack vector that can be exploited for malicious purposes, researcher Flesch argues in his description. Moreover, he says, the vulnerability points to poor programming and a lack of attention to security.
In his argument, the security researcher advises OpenAI to solve this problem quickly. The AI giant could do this by setting strict limits on the number of URLs included in a request. Filtering for duplicate requests should also be added, and other restrictive measures should be taken to prevent abuse.
Also read: ChatGPT search proves highly susceptible to manipulation