This stack sits on an authentication path: it stores fingerprint templates and speaks the D-Bus contract a PAM login trusts. The pure crates are #![forbid(unsafe_code)] and the arithmetic kernels are verified bit-exact against stock NBIS, but a logic flaw here is an authentication flaw — so we take security reports seriously.

Please do not open a public issue for security problems.

Report privately through GitHub's private vulnerability reporting (Security → Advisories → Report a vulnerability). Include:

• the affected crate and version / commit,
• the platform and how the daemon was configured, and
• a minimal reproducer and the observed impact.

We aim to acknowledge a report within a few days and will keep you updated as we investigate. Once a fix is available we will coordinate a disclosure timeline with you and credit you unless you prefer otherwise.

In scope:

• Template handling and storage — the FP3 codec (fprint-fp3) parsing untrusted or attacker-influenced template bytes, and the /var/lib/fprint layout the daemon writes (permissions, path handling, cross-user leakage).
• The D-Bus surface (fprintd) — PolicyKit action enforcement, client isolation, and any way to bypass or confuse a verify/identify result.
• The FFI boundary in the shim (fprint-backend-libfprint) and unsafe in any transport leaf.
• The arithmetic kernels (fprint-bozorth3, fprint-mindtct) — panics, unbounded memory/CPU, or match scores that diverge from NBIS in a way affecting security.

Out of scope:

• The experimental USB capture seam (fprint-backend-native's usb module). Its protocol values are unverified placeholders and it cannot talk to real hardware; it is a worked example, not a shipping driver.
• Weaknesses inherent to fingerprint biometrics as an authentication factor (presentation attacks, sensor spoofing) rather than to this implementation.

Pre-1.0: only the latest main receives security fixes.

The shim daemon has so far been exercised only against libfprint's virtual drivers in Docker; it has not been hardened or reviewed against a real sensor or a real PAM login. Treat current revisions accordingly.