7 ChatGPT Bugs Allow Rampant Data Theft

Other than allowing a hacker to inject arbitrary prompts, exfiltrate personal user data, bypass safety mechanisms and take other malicious actions, the collection of bugs do not represent a major problem for users. Needless to say, I am kidding about the last part. It could be a major problem.

The exfiltration steals data from your chat history and stored memories.

The bugs mostly stem from how ChatGPT and SearchGPT behave when a user is browsing the web. The allow the software to manipulate the chatbot’s behavior is ways the user cannot detect.

The researchers, at security firm Tenable, said:

“By mixing and matching all of the vulnerabilities and techniques we discovered, we were able to create proofs of concept (PoCs) for multiple complete attack vectors,” 

https://www.darkreading.com/application-security/multiple-chatgpt-security-bugs-rampant-data-theft

This is neither the first time nor the last time that we will see security and privacy bugs in LLM software. No different that patching Windows or the Chrome browser every month. Just something we have to live with.

In a nutshell, the seven vulnerabilities Tenable uncovered stem from how ChatGPT ingests and processes instructions from external sources, including websites it browses, search results, blog comments, and specially crafted URLs. The security vendor showed how attackers could exploit the flaws by hiding malicious prompts in blog comments, poisoning search results to bypass ChatGPT’s safety filters and taking advantage of how ChatGPT processes conversation history and stores memories. 

One of the bugs allows an adversary to plant malicious instructions on a trusted web page because of how the bot interacts with the website. There are multiple ways the attacker could launch an attack. Some require a little bit of user interaction; others do not.

While they did their research on ChatGPT-4o, they discovered at at least some of the bugs, like the prompt injection attack, also work on ChatGPT-5.

Given how popular AI is, attackers (as well as researchers) are going to be working overtime looking for bugs. And, I am sure they will find them.

Credit: Dark Reading

by