TLDR AI · MISCELLANEOUS

EXTENSIONS PROMPT INJECTION AND DATA EXFILTRATION

embracethered.com12 min read

Summary from the TLDR AI newsletter

A great deep dive into some novel vulnerabilities that come from generative AI plugins. In this case, the attack resembles SQL injection but came via an infected Google Doc. While Google has fixed the issue, the security researcher was able to exfiltrate other users’ prompts.