Please do not open a public issue for a security problem.
Report it privately through GitHub: Report a vulnerability (Security → Advisories → Report a vulnerability). That opens a private thread visible only to you and the maintainers.
Helpful things to include, when you have them:
- Which front-end and version (
fileid --version, or Settings → Engine info) - OS and architecture
- A minimal reproduction — a crafted file, a command, or a sequence of steps
- What an attacker gains, and whether it needs local access or a user action
Reports are read as time allows; this is a small project, so please don't expect same-day turnaround. You'll get an acknowledgement, and a heads-up before any advisory naming you is published. If you'd rather not be credited, say so.
FileID is pre-1.0 and ships as unsigned prereleases. Only the latest release
and main receive fixes — there are no backports to older tags.
| Version | Supported |
|---|---|
main |
✅ |
| Latest release | ✅ |
| Anything older | ❌ |
FileID runs entirely on-device. There is no FileID server, account, or hosted
API, so there is no remote attack surface to report against. The interesting
boundaries are local, and are documented in detail — including the threat model,
what's enforced today, and what's still open — in
shared/docs/SECURITY.md.
Things that are in scope:
- Malicious file content that exploits the scan pipeline (image/PDF/archive decoders, OCR, face detection, the VLM, hashing)
- Escaping the Restructure/rename path containment, or otherwise getting FileID to write, move, or delete outside the user-selected library root
- Defeating the engine-binary integrity check the app performs before spawn
- Tampering with a model or runtime download (transport, digest, or redirect handling)
- Any network egress beyond a user-initiated model download — the project ships
no telemetry and treats a violation as a release blocker
(see
shared/docs/PRIVACY.md)
Things that are out of scope:
- The absence of Authenticode/notarization signatures on prerelease builds —
this is known, documented, and tracked in
shared/docs/SHIP.md - Vulnerabilities in third-party model weights or in tools you supply on
PATH(llama-mtmd-cli,tesseract,ffmpeg, HEIC converters). These sit inside your local trust boundary — report those upstream. - An attacker who already has code execution as your user account, or administrator/root on the machine