Well documentedThree or more independent publishers report the core facts.
Security researcher demonstrates ChatGPT memory exploit for data exfiltration
A security researcher publicly demonstrated how attackers could exploit ChatGPT's memory feature via prompt injection to plant false memories and gain persistent access to user data. OpenAI has reportedly fixed the vulnerability.
AII-1002-459616 sources ↓
What happened
A security researcher demonstrated a vulnerability in ChatGPT's memory feature on the macOS app that could allow an attacker to inject false memories into a user's long-term storage through indirect prompt injection.
Reported impact
The vulnerability could enable attackers to store malicious instructions and establish a persistent exfiltration channel to access all of a user's input indefinitely.
Governance controls
OpenAI has fixed the flaw.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.
Ars Technica
False memories planted in ChatGPT give hacker persistent exfiltration channel
24 September 2024
BGR
ChatGPT memory exploit left your private chat data exposed, but OpenAI fixed it
25 September 2024
WinBuzzer
OpenAI Faces Security Challenge with ChatGPT Memory Vulnerability - WinBuzzer
25 September 2024
TechRadar
Hacker creates false memories in ChatGPT to steal victim data -- but it might not be as bad as it sounds
26 September 2024
TechSpot
ChatGPT vulnerability allows hackers to record your sessions indefinitely
26 September 2024