OpenAI’s ChatGPT has introduced a new “remember” feature, allowing the AI chatbot to retain details from conversations and store them in a memory file. Although this feature aims to improve ChatGPT’s efficiency by providing a personalized experience, security concerns have arisen regarding its potential misuse. Johann Rehberger, a security researcher, recently uncovered vulnerabilities in ChatGPT’s “long-term conversation memory” feature, highlighting how it can be easily manipulated.
The feature was initially rolled out in a beta version in February and has since become widely available. Rehberger demonstrated how the memory feature could be misused by providing ChatGPT with false information through a Microsoft Word document. He successfully convinced ChatGPT to store misleading details as facts, such as making the AI believe that he was over 100 years old and lived in the Matrix, a fictional universe.
Rehberger’s research revealed a critical flaw in the system: not only can ChatGPT be manipulated into storing incorrect data, but it could also transfer this false information to external servers. Initially, OpenAI did not treat this as a security risk, dismissing it as a “Model Safety Issue.” However, once Rehberger provided clear evidence, OpenAI patched the system to prevent data from being sent to third-party servers. Despite this fix, the issue of memory manipulation remains unresolved.
In his demonstration video, Rehberger showed how ChatGPT could still be manipulated to store incorrect information through external sources like websites or untrusted documents. This vulnerability poses a significant concern, as the chatbot’s long-term memory could be used to spread misinformation or cause harm by tricking it into storing and sharing false data.
As of now, OpenAI has not fully addressed the memory manipulation flaw, and the public is waiting for further updates on how the company plans to resolve this issue. The new memory feature, while promising in enhancing user interaction, exposes the system to potential risks, raising important questions about AI security and responsible usage.
Get Latest News Live on Times Now along with Breaking News and Top Headlines from Technology Science and around the world.
