Researchers were able to hack ChatGPT memories.

Researchers uncovered seven vulnerabilities in ChatGPT, including flaws in its memory and web browsing features that allow for prompt injection and data exfiltration. Attackers can exploit SearchGPT’s behavior to execute hidden prompts from websites or use Bing URLs to bypass safety checks and steal user data. Some methods involve injecting malicious instructions into websites or tricking users into clicking crafted links that execute commands automatically.

Read more:

https://www.securityweek.com/researchers-hack-chatgpt-memories-and-web-search-features/