We recommend the following best practices for managing role recommendations.

For more information about role recommendations, see the
[role recommendation overview](https://docs.cloud.google.com/policy-intelligence/docs/role-recommendations-overview.md.txt).

## Getting started with recommendations

The following best practices can help you get started with role recommendations.

- **Begin with an initial cleanup of over-granted permissions.** Initially, you
  might see a very large number of recommendations, especially if many
  principals have highly permissive roles like Editor. Take the time to address
  all recommendations in your project or organization to ensure that all of your
  principals have the appropriate roles.

  When doing this initial cleanup, prioritize the following types of
  recommendations:
  - **Recommendations that reduce permissions for service accounts.**
    By default, all [default service accounts](https://docs.cloud.google.com/iam/docs/service-account-types.md.txt#default) are
    granted the highly permissive Editor role on projects. Other
    service accounts that you manage might also have been granted highly
    permissive roles. All over-granted permissions increase your security
    risk, including overly privileged service accounts, so we recommend
    prioritizing overly privileged service accounts during your
    initial cleanup.

  - **Recommendations that help prevent privilege escalation.** Roles that
    let principals act as a service account (`iam.serviceAccounts.actAs`)
    or get or set the allow policy for a resource can potentially let a
    principal escalate their own privilege. Prioritize recommendations
    relating to these roles.

  - **Recommendations that reduce lateral movement.** Lateral movement is when
    a service account in one project has permission to impersonate a service
    account in another project. This permission can result in a chain of
    impersonations across projects that gives principals unintended access to
    resources. To mitigate this unintended access, prioritize recommendations
    that are associated with [lateral movement
    insights](https://docs.cloud.google.com/policy-intelligence/docs/role-recommendations-overview#lateral-movement-insights).

  - **Recommendations with a high priority level.** IAM
    recommendations are automatically assigned priority levels based on the
    role bindings they're associated with. Prioritize recommendations with
    a high priority level to quickly reduce over-granted permissions.

    To learn how a recommendation's priority is determined, see
    [Recommendation priority](https://docs.cloud.google.com/policy-intelligence/docs/role-recommendations-overview.md.txt#priority).
  - **When you find an over-privileged principal in one project, check other
    projects for recommendations involving that principal.** If a principal
    has been granted an overly permissive role in one project, it is possible
    that they have been granted overly permissive roles in other projects as
    well. Review recommendations for the principal across multiple projects to
    globally reduce the principal's access to the appropriate level.

- After the initial cleanup, **check your recommendations regularly.** We
  recommend that you check your recommendations at least once a week. This check
  will usually take much less time than the initial cleanup, because you will
  only need to address recommendations for changes that have occurred since the
  last cleanup or check.

  Regularly checking permissions reduces the work required for each check, and
  can help you proactively identify and remove inactive users, as well as
  continue to downscope permissions for active users.

## Best practices for working with recommendations

If you use the [Recommender API](https://docs.cloud.google.com/recommender/docs/reference/rest.md.txt) or the
[`recommender` commands for the gcloud CLI](https://docs.cloud.google.com/sdk/gcloud/reference/recommender) to
manage recommendations, make sure to update the state of recommendations that
you apply. This allows you to keep track of your recommendations and ensures
that the changes you make appear in your [recommendations logs](https://docs.cloud.google.com/policy-intelligence/docs/review-apply-role-recommendations.md.txt#logs).

Before you apply a recommendation, review it carefully and make sure you
understand when it was last refreshed and how it will change the principal's
access to Google Cloud resources.

## Best practices for applying recommendations automatically

To manage your recommendations more efficiently, you might want to automate the
process of applying recommendations. If you decide to use automation, keep the
following points in mind.

Recommender tries to provide recommendations that will not
cause breaking changes in access. For example, we will never recommend a role
that excludes permissions that a principal has used,
[passively or actively](https://docs.cloud.google.com/policy-intelligence/docs/role-recommendations-overview.md.txt#permissions-used), in the last
90 days. We also use [machine learning](https://docs.cloud.google.com/policy-intelligence/docs/role-recommendations-overview.md.txt#ml) to
identify other permissions that the user is likely to need.

However, we cannot guarantee that our recommendations will never cause breaking
changes in access---it is possible that applying a recommendation will
result in a principal being unable to access a resource that they need. We
recommend reviewing
[How the IAM recommender works](https://docs.cloud.google.com/policy-intelligence/docs/role-recommendations-overview.md.txt#how-recommender-works) and
deciding how much automation you are comfortable with. For example, you might
decide to apply most recommendations automatically, but require a manual review
for recommendations that add or remove a certain number of permissions, or that
involve granting or revoking a specific role.

When automating recommendations, you might want to identify which resource a
recommendation is for. To identify the resource, use the `operation.resource`
field. Other fields, such as the `name` field, will not always represent the
resource that the recommendation is for.

## What's next

- Understand [role recommendations](https://docs.cloud.google.com/policy-intelligence/docs/role-recommendations-overview.md.txt).
- Learn the steps for [reviewing and applying recommendations](https://docs.cloud.google.com/policy-intelligence/docs/review-apply-role-recommendations.md.txt).
- Find out how to [export data for IAM recommendations](https://docs.cloud.google.com/policy-intelligence/docs/export-role-recommendations-data.md.txt).