Guide, 10 Nov 2026, read 10 Nov 2026
Unused access is a count of grants; excess reachability is a count of paths, and they need different fixes
Unused access measures the gap between granted and used
Most CIEM tools start here because the data is available. Google documents that the IAM recommender compares a principal's permissions with those it used in the last 90 days and recommends removing or replacing the role. AWS documents that IAM Access Analyzer unused access findings highlight unused roles, unused access keys and unused passwords, and for active roles show unused services and actions.
Microsoft's retired Entra Permissions Management summarized the same idea as a number: its Permission Creep Index compared permissions granted with permissions exercised. The method is simple and defensible, since a permission nobody has used in the window is a candidate for removal. Its limit is that it tells you nothing about which of the used permissions are dangerous. A count of this kind says how much was granted and went unused. It says little about how much damage the used part could do.
Excess reachability measures what can be reached
Reachability is a different question. Act's engineering blog states it as two tests that every cloud action must pass at once: the network must let the caller reach the resource, and IAM must permit the action. Act argues the two checks are run by separate teams in separate tools, so nobody computes what a compromised workload can reach with both fused. That is Act's framing.
Microsoft Defender for Cloud describes the same shape from the attacker side. Its documentation defines an attack path as a series of steps that starts at an external entry point, such as an internet-exposed vulnerable resource, and follows lateral movement through attached identities with permissions to other resources. The unit is a path, and a path has several links. Each link in such a path is ordinary on its own, and the chain is what an attacker follows.
The two measures disagree in both directions
An unused permission on an identity nothing can reach is low risk and still counts against a usage metric. A permission used daily by a workload that is exposed to the internet, through a role that can read a production database, never appears as unused and is the one an attacker wants. Wiz's CIEM guidance reaches a similar view in its own words: effective permissions are the reliable signal, and CIEM should prioritize identity-based attack paths over raw permissions.
Neither measure replaces the other. Usage data answers what can be removed with little argument. Reachability answers which removals matter first and which boundary removes the most paths in one change. Run the usage review first because it is quick and its results are easy to defend. Then use reachability to order the larger changes that follow, so the cheap work does not crowd out the work that lowers risk.
The fixes differ too
Unused access is removed grant by grant: delete the role, shrink the policy to the generated template, rotate out the key. AWS describes policy generation from up to 90 days of CloudTrail activity for that purpose. The change is small, local and easy to simulate against the same logs. One engineer can usually run that clean-up in an afternoon, and the same engineer can check the result against the logs.
Excess reachability is usually removed at a layer above the grant: a network rule, a trust condition, or an organization-level policy that denies a class of movement. Act describes guardrails that deny whole classes of cross-environment movement and ring-fencing, with explicit allow, around crown-jewel workloads. Sonrai describes organization-level policies that block unused privileges. These are vendor claims; the method shows how this index scores them.
A program should track both and say which one it is moving
Report the two numbers separately. Unused grants removed per month shows hygiene progress and is cheap to produce. Paths removed, or the reach of the top ten crown-jewel resources, shows risk progress. Reachability needs a model that joins network and identity, which is a larger purchase and a larger project than a usage report, so most teams begin with the first number and grow into the second.
When a vendor says it measures least privilege, ask which of the two it means. A tool built on usage will show a falling count of unused permissions while the reach of a compromised role is unchanged. See reading a vendor least privilege claim for the questions to put to each vendor. Keep the vendor's answers beside its platform profile, with the date.
Sources
- Google Cloud, overview of role recommendations, read 10 Nov 2026
- AWS, what is IAM Access Analyzer, read 10 Nov 2026
- Microsoft Learn, Entra Permissions Management FAQ, read 10 Nov 2026
- Act blog, network and identity each see half your cloud, read 10 Nov 2026
- Microsoft Learn, attack paths in Defender for Cloud, read 10 Nov 2026
- Wiz Academy, how to choose a CIEM vendor, read 10 Nov 2026
- AWS IAM Access Analyzer policy generation, read 10 Nov 2026
Last reviewed 10 Nov 2026.