Google Gemini Download Feature Reportedly Delivered Images Its Safety Filters Had Blocked
AI SECURITYGoogleredteam/daily · 2 sources
- Severity
- Medium
- Sources
- 2
What happened
Fog Security published research on 5 November 2026 showing that Gemini's full-size download path could return images that the chat interface had refused, while the displayed preview was a permitted image. No jailbreak was needed, and no access to other users' data or privilege escalation was reported. Fog reported it in June 2026, and by 22 October it could no longer reproduce the issue in 12 attempts, but Google attributed that to routine maintenance and closed the report on 23 October without an internal ticket.
For IT people · Written by AI from 2 sources · Wrong?
What to do
- AI developers should bind each download to its approved preview and remove files from refused or failed generations, as Fog Security recommends.
- Test AI safety controls across stored files and download links, not only chatbot replies.
Written by AI from the sources, checked against them
Full story timeline · 2 updates
Who reported · 2 outlets
GBHackersCyber Security News
Who and what is in it
Companies
Products