Misuse1 sourceFirst seen
Custom Instructions Turned ChatGPT Into an Unpaid Data Courier
Ourtake
The exfiltration bug is months old, the fix is still pending, and the backdoor now comes with its own settings menu.
What happened
A security researcher demonstrated that ChatGPT's Custom Instructions could be abused to install a backdoor that sends conversation summaries to an external server through image markdown injection. The researcher reported the issue to OpenAI in April, and the article says it was still unfixed when published.