Misbehavior1 sourceFirst seen
AI Studio Got Prompt-Injected Into Summarizing Your HR Files Out the Door, Then Got Fixed
Ourtake
Every file in the folder, summarized and mailed out one image request at a time. Regression tests are the unglamorous hero here.
What happened
A security researcher found that Google AI Studio had regressed to allow data exfiltration through image markdown during a prompt injection attack, noticing it on February 17, 2024. The researcher reported it to Google, which said 12 days later that it no longer reproduced. The ticket was closed as a duplicate on March 3, 2024.