Executive brief
prompts.chat, a platform for sharing and managing AI prompts, contained a security flaw that allowed unauthorized users to view private content. Attackers could bypass privacy settings to access sensitive prompt history, change requests, and internal metadata that should have been restricted to owners and administrators. This could lead to the exposure of proprietary or confidential AI instructions and user-submitted examples.
Technical details
The vulnerability stems from a missing 'isPrivate' check across several API endpoints and the metadata generation logic in prompts.chat. Specifically, endpoints for version history, change requests, examples, and voting failed to verify if a prompt was marked as private before returning data. An unauthenticated remote attacker could exploit these omissions to retrieve sensitive information, including prompt content, version history, and user metadata exposed via HTML meta tags. The issue was addressed in commit 7b81836 by implementing a centralized 'checkPromptAccess' guard that enforces privacy logic based on the prompt's status and the user's session.
Affected products
- fka prompts.chat prior to commit 7b81836
Timeline
- 2026-03-25: patched: Fix committed to repository
- 2026-04-03: disclosed: Initial advisory published