Memory Data Hording

Type: technique

Description: The adversary gets AI to commit malicious text into memory that instructs it to retain sensitive data in its memory in future threads, for future exfiltration.

Version: 0.1.0

Created At: 2024-10-11 16:54:32 +0300

Last Modified At: 2024-10-11 16:54:32 +0300

External References

  • --> Collection (tactic): An adversary that successfully infected one thread can get the AI system to collect information from other threads.
  • <-- Exfiltration of personal information from ChatGPT via prompt injection (procedure): Assuming we can execute 2 prompt injections, separated by some period of time (e.g., a month), we can use ChatGPT's memory feature to remember the desired value using the first prompt and extract it using the second prompt.