Gemini's Memory Gets Gaslit by a Document It Was Only Asked to Summarize

What happened

A security researcher showed that a document with hidden instructions could make Google's Gemini store false information in a user's long-term memory. The attack used delayed tool invocation, where the injected instruction ran only after the user replied with a trigger word such as 'yes'. The researcher reported the issue to Google, which assessed it as low likelihood and low impact.