# Read a least privilege claim by asking who applies the change and what it was tested against

Guide, 10 Nov 2026, read 10 Nov 2026

**The short answer.** Vendors use "least privilege" for four different outcomes: a report, a recommendation, a policy you apply, or a policy the product enforces. Five questions separate them: who applies the change, what tested it, what data window it used, which access it models, and what proof remains afterward.

## The same phrase covers four outcomes

Provider documentation shows the range. AWS documents that IAM Access Analyzer generates a policy template from CloudTrail activity and that you create and attach the policy yourself. Google documents that the IAM recommender does not apply recommendations automatically. Those are recommendations, and both providers say so plainly. The documentation of each provider and vendor is a better guide than the headline claim, and that is where the questions below point.

Product pages are less plain. A sentence such as "automatically enforces least privilege" can mean the product blocks access itself, or that it writes a policy for you to deploy. Both can be accurate, and they differ in who owns the last step and the outage risk. Read every claim as a claim by that vendor, and look for the page that describes the mechanics.

## Question one: who applies the change

This is the largest weight on the index, 25 percent. Look for the verb on the documentation page, not the home page. Sonrai's platform page says unused privileged permissions are blocked with cloud-native, org-level policies, while another Sonrai page says permissions unused for a time are suggested for removal and not removed automatically. Both statements are Sonrai's. Ask also which party takes the outage if the change is wrong.

Act says it simulates a change and the customer applies it through the cloud's native controls. Act has read-only access, and its changes arrive as pull requests to Terraform or CloudFormation, or as tickets. Neither model is wrong. They place the final step with a different party. Rollback belongs to the same question: Act's page says changes can be rolled back, while an analyst report lists rollback among items still under discussion.

## Questions two and three: what tested it, and over what window

Ask whether each change is simulated against the real environment and whether the impact is shown. Google documents a replay of up to 90 days of access logs. Act says it tests against a digital twin and offers audit mode. Sonrai says in-use permissions are exempted first and the list is shown before deployment. A vendor that gives the window and the failure cases is giving you what you need to judge the claim.

Then ask for the observation window. A tool that sees 30 days will mark a quarterly job's permissions as unused. Google's own documentation warns that shortening the minimum observation period gives recommendations sooner but might affect accuracy. The vendor should state its window as plainly as Google does. Act's materials describe an audit mode for this purpose, and the analyst report says audit mode uses audit-log events observed against the policy.

## Questions four and five: which access it models, and what proof remains

Coverage means identity, network and AI access in one model. A usage-based tool covers identity. A tool that joins network and IAM covers reachability. Ask whether AI workloads and agents appear in the model or only in a roadmap slide. An analyst report on Act (SACR) says customer evidence for Act's AI workload model is not yet as mature as for its broader access control.

Proof means what an auditor can read afterward: the policy as code, the simulation result, the approval and the date it shipped. Ask how exceptions are handled in the team's own tools and whether violations are caught before production. The same SACR report says Act identifies policy drift and presents violations management and time-bound waiver requests as near-term roadmap. Treat roadmap items as absent until they ship, and date the note.

## A reading worksheet

Put each claim in a row and fill the columns from documentation, not from the sales call. A blank cell is a finding: write "none found" with the date you looked. The table below is the worksheet in compact form, with the weak answer for each row. Keep the source next to each row, so a later reader can see which page supported which answer.

The platform profiles on this index follow the same rule. Each one lists what the pages read say, what they do not say, and the date they were read. See [the method](/method/) for the seven criteria and [Sonrai Security](/platforms/sonrai/) or [Act](/platforms/act/) for two worked profiles. Both profiles name their own limits next to their strengths. The scores in the main view are the result of asking these questions of every vendor on the same date.

| Question | Where to look | Weak answer |
| --- | --- | --- |
| Who applies the change | Documentation verb: blocks, recommends, opens a pull request | Only "automates" or "enforces" |
| What tested it | Simulation or audit mode page | No test before enforcement |
| What window | Observation period setting | Window not stated |
| Which access | Identity, network, AI model | Identity only |
| What proof | Policy as code, logs, approvals | A dashboard screenshot |

## Sources

- AWS IAM Access Analyzer policy generation: https://docs.aws.amazon.com/IAM/latest/UserGuide/access-analyzer-policy-generation.html (read 10 Nov 2026)
- Google Cloud, overview of role recommendations: https://cloud.google.com/policy-intelligence/docs/role-recommendations-overview (read 10 Nov 2026)
- Sonrai Security, platform page: https://sonraisecurity.com/platform (read 10 Nov 2026)
- Sonrai Security, least privilege use case: https://sonraisecurity.com/use-cases/least-privilege (read 10 Nov 2026)
- Act Security, platform page: https://act.security/platform (read 10 Nov 2026)
- Google Cloud, Policy Simulator for allow policies: https://cloud.google.com/policy-intelligence/docs/iam-simulator-overview (read 10 Nov 2026)
- SACR, the CISO's guide to cloud security intent enforcement: https://softwareanalyst.io/reports/the-ciso-s-guide-to-cloud-security-intent-enforcement (read 10 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://cloudleastprivilege.com/guides/reading-a-vendor-least-privilege-claim/
