Comparing Centrally and Locally Verified Memorized Secrets

Jim Fenton, a consultant on the NIST SP 800-63 revision, opens by separating memorized secrets used to log in to a central service from activation secrets. An activation secret is the PIN or passphrase a user types to unlock a multi-factor authenticator, such as a hardware token, smart card or phone-based credential. He argues the earlier draft lumped the two together, which fit poorly, because the threats differ. Instead of cracking a breached server database, an attacker of an activation secret is more likely to combine shoulder surfing with theft of the device, or to use side channels on the hardware.

The talk then walks through design factors: length and composition, and the fact that short PINs cluster around dates and years, so extra digits add less than expected. Block lists of common values help only modestly, because attackers can see the list and simply skip those values. Throttling is central, but many standalone authenticators have no clock, so they fall back to hard lockouts after around ten failed tries. That can brick the device, which makes a backup authenticator essential. Biometrics came up in the Q&A as a complement rather than a replacement, since sensors fail with wet hands or masks.

Fenton's takeaway is that activation secrets need their own guidance, and that password rules such as long required lengths do not transfer, since many authenticators accept only a few characters. Practitioners should check each authenticator's retry limit, secret size and storage limits before relying on it. He noted that the next draft of the guidance was still subject to review.