2.2.2. Microsoft Entra Password Protection
💡 First Principle: Attackers don't guess passwords at random — they try the passwords people actually choose: the company name with a year, the season with an exclamation mark. Password Protection blocks the patterns attackers try first, which is why it normalizes and fuzzy-matches rather than only comparing exact strings.
Password Protection evaluates every password change or reset against two lists:
- The global banned password list, maintained by Microsoft from attack data. It applies to everyone automatically, can't be configured or disabled, and isn't published.
- The custom banned password list, up to 1,000 terms, where you add your organization's own weak terms — brand names, product names, office locations, sports teams. Use base terms; you don't need to list variations.
Evaluation normalizes the candidate password (lowercasing, and substitutions such as 0→o, 1→l, $→s, @→a), fuzzy-matches banned terms at an edit distance of one, and substring-matches the user's first and last name and the tenant name. Each banned term found scores one point and each remaining character scores one point; a password needs at least five points to be accepted. So in Contoso's tenant C0nt0s0!! normalizes to contoso!!: one point for the banned term plus two for the remaining characters makes three, and it's rejected, even though no list contains that exact string. Adding length raises the score — Contoso2026! scores six and passes — so the custom list stops the weakest variants rather than every password containing a banned word.
Licensing:
| Users | Global list | Custom list |
|---|---|---|
| Cloud-only | Free | Entra ID P1 or P2 |
| Synchronized from on-premises AD DS | P1 or P2 | P1 or P2 |
On-premises. To apply the same lists to passwords changed on domain controllers, you install the DC agent on domain controllers and a proxy service that fetches the policy from Entra. On-premises enforcement can run in Audit mode (log what would be rejected) or Enforced mode; audit mode exists only on-premises — the cloud always enforces.
Smart lockout sits alongside it, locking out sign-in attempts that look like attacks while letting the real user in. Defaults are a threshold of 10 failed attempts and a 60-second lockout; customizing them needs P1. In a hybrid environment, set the Entra lockout threshold lower than the AD DS threshold so Entra locks out the attacker before the on-premises account is locked.
⚠️ Exam Trap: "Test the impact of banned passwords before enforcing them" is possible only for on-premises domain controllers, through the DC agent's Audit mode. There is no audit mode for cloud password changes.
Reflection Question: Contoso adds "contoso" to its custom banned list. A user tries C0nt0s0!!. Walk through its score and why it's rejected, then explain why a longer password that still contains "contoso" might pass.