Executive brief
Banana Slides, an AI-powered presentation generator, contains a security flaw that allows unauthorized users to access private files on the server. By including specially crafted image links in a presentation description, an attacker can trick the system into reading image files from folders they should not have access to. This could lead to the exposure of sensitive images or data stored in backup or temporary directories on the host system.
Technical details
A path traversal vulnerability exists in the `generate_image()` function within `backend/services/ai_service.py`. The application attempts to restrict file access to the `uploads` directory using `os.path.startswith()`, but fails to include a trailing path separator in the prefix check. An unauthenticated attacker can exploit this by providing markdown image references (e.g., ``) that resolve to sibling directories sharing the same name prefix as the upload folder. When processed, the application uses `PIL.Image.open()` to read the file from the unauthorized location. The issue is patched in commit `e8bc490` by appending `os.sep` to the validation prefix.
Affected products
- Anionex Banana Slides through 0.4.0
Timeline
- 2026-05-31: patched: Fix merged in commit e8bc490
- 2026-06-01: disclosed: Vulnerability disclosed via GitHub Issue 429 and VulnCheck advisory
- 2026-06-01: advisory: CVE-2026-49136 published to NVD