Azure AD - Fuzzy Policy

In Azure AD Password Protection, there exists a system policy called "Fuzzy Policy". The policy checks user password reset request and verifies the suggested password against a global banned password list.

The Azure AD Identity Protection team constantly analyzes Azure AD security telemetry data looking for commonly used weak or compromised passwords. Specifically, the analysis looks for base terms that often are used as the basis for weak passwords. When weak terms are found, they're added to the global banned password list. The contents of the global banned password list aren't based on any external data source, but on the results of Azure AD security telemetry and analysis.

When a password is changed or reset for any user in an Azure AD tenant, the current version of the global banned password list is used to validate the strength of the password. This validation check results in stronger passwords for all Azure AD customers.

The global banned password list is automatically applied to all users in an Azure AD tenant. There's nothing to enable or configure, and can't be disabled. This global banned password list is applied to users when they change or reset their own password through Azure AD.

How are passwords evaluated

When a user changes or resets their password, the new password is checked for strength and complexity by validating it against the combined list of terms from the global and custom banned password lists.

Even if a user's password contains a banned password, the password may be accepted if the overall password is otherwise strong enough. A newly configured password goes through the following steps to assess its overall strength to determine if it should be accepted or rejected:

Step 1: Normalization

A new password first goes through a normalization process. This technique allows for a small set of banned passwords to be mapped to a much larger set of potentially weak passwords.

Normalization has the following two parts:

• All uppercase letters are changed to lower case.

• Then, common character substitutions are performed, such as in the following example:

  TABLE 1

  Original letter	Substituted letter
  0	o
  1	l
  $	s
  @	a

Consider the following example:

• The password "blank" is banned.
• A user tries to change their password to "Bl@nK".
• Even though "Bl@nk" isn't banned, the normalization process converts this password to "blank".
• This password would be rejected.

Step 2: Check if password is considered banned

A password is then examined for other matching behavior, and a score is generated. This final score determines if the password change request is accepted or rejected.

Fuzzy matching behavior

Fuzzy matching is used on the normalized password to identify if it contains a password found on either the global or the custom banned password lists. The matching process is based on an edit distance of one (1) comparison.

Consider the following example:

• The password "abcdef" is banned.

• A user tries to change their password to one of the following:

  • 'abcdeg' - last character changed from 'f' to 'g'
  • 'abcdefg' - 'g' appended to end
  • 'abcde' - trailing 'f' was deleted from end

• Each of the above passwords doesn't specifically match the banned password "abcdef".

  However, since each example is within an edit distance of 1 of the banned term 'abcdef', they're all considered as a match to "abcdef".

• These passwords would be rejected.

Substring matching (on specific terms)

Substring matching is used on the normalized password to check for the user's first and last name as well as the tenant name. Tenant name matching isn't done when validating passwords on an AD DS domain controller for on-premises hybrid scenarios.

Consider the following example:

• A user named Poll who wants to reset their password to "p0LL23fb".
• After normalization, this password would become "poll23fb".
• Substring matching finds that the password contains the user's first name "Poll".
• Even though "poll23fb" wasn't specifically on either banned password list, substring matching found "Poll" in the password.
• This password would be rejected.

  
  Important

  Substring matching is only enforced for names, and other terms, that are at least four characters long.
  
  
  

  Score Calculation

  The next step is to identify all instances of banned passwords in the user's normalized new password. Points are assigned based on the following criteria:

  1. Each banned password that's found in a user's password is given one point.
  2. Each remaining character that is not part of a banned password is given one point.
  3. A password must be at least five (5) points to be accepted.

  For the next two example scenarios, Contoso is using Azure AD Password Protection and has "contoso" on their custom banned password list. Let's also assume that "blank" is on the global list.

  In the following example scenario, a user changes their password to "C0ntos0Blank12":

  • After normalization, this password becomes "contosoblank12".

  • The matching process finds that this password contains two banned passwords: "contoso" and "blank".

  • This password is then given the following score:

    [contoso] + [blank] + [1] + [2] = 4 points

  • As this password is under five (5) points, it's rejected.

  Let's look a slightly different example to show how additional complexity in a password can build the required number of points to be accepted. In the following example scenario, a user changes their password to "ContoS0Bl@nkf9!":

  • After normalization, this password becomes "contosoblankf9!".

  • The matching process finds that this password contains two banned passwords: "contoso" and "blank".

  • This password is then given the following score:

    [contoso] + [blank] + [f] + [9] + [!] = 5 points

  • As this password is at least five (5) points, it's accepted.

  Important

  The banned password algorithm, along with the global banned password list, can and do change at any time in Azure based on ongoing security analysis and research.

  For the on-premises DC agent service in hybrid scenarios, updated algorithms only take effect after the DC agent software is upgraded.

• What do users see

  When a user attempts to reset a password to something that would be banned, the following error message is displayed:

  "Unfortunately, your password contains a word, phrase, or pattern that makes your password easily guessable. Please try again with a different password."

  License requirements

  LICENSE REQUIREMENTS

  Users	Azure AD Password Protection with global banned password list	Azure AD Password Protection with custom banned password list
  Cloud-only users	Azure AD Free	Azure AD Premium P1 or P2
  Users synchronized from on-premises AD DS

  Note

  On-premises AD DS users that aren't synchronized to Azure AD also benefit from Azure AD Password Protection based on existing licensing for synchronized users.

  Microsoft Documentation: https://docs.microsoft.com/en-us/azure/active-directory/authentication/concept-password-ban-bad#how-are-passwords-evaluated