Simulate a permission change against real access logs, then read what the simulator cannot see
Facts read 10 Nov 2026. Page reviewed 16 Nov 2026.
A permission change can be tested before it ships by replaying past access attempts under the proposed policy. Google Policy Simulator does this with 90 days of logs, AWS offers a policy simulator, and Act describes a digital twin. Each has stated limits that decide how far to trust it.
A simulation compares outcomes, not policy text
Diffing two policy files shows what changed in the text. It does not show who loses access. A useful simulation takes real or representative requests and evaluates each under the current and the proposed policy, then reports the requests whose result differs. Those differences are the people and workloads you would break. Any simulator worth using reports exactly this list, whether a provider ships it or a vendor builds it.
Google documents this method for allow policies. Policy Simulator retrieves access logs from the last 90 days, replays the access attempts under the current policy, replays them again under the proposed policy, and reports the differences as a list of access changes. Because it replays under the current policy first, it reports only changes caused by your proposal. Those two replays are what separate a simulation from a comparison of policy documents.
Simulation is also what lets a team say yes to a removal. Without it the choice is between guessing and waiting, and waiting is what stalled the project in the first place. With it, each change comes with a list of the access it would have changed, which is the evidence a reviewer needs. The reviewer's job becomes reading differences, which is faster and safer than reading policy text.
The provider tools state their limits
Read the limits before you trust a result. Google documents that Policy Simulator for allow policies fails if there are more than 5,000 access logs to replay for the project or organization, and that it lists permissions on unsupported resource types as ones it could not simulate. Each Google simulator evaluates one policy type and does not account for the others.
AWS documents that its IAM policy simulator tests identity-based policies, permissions boundaries, service control policies and resource-based policies that you provide, without sending a real request. AWS also warns that results can differ from your live environment and recommends checking against the live environment afterward. A simulator that tests supplied policies is checking logic, and the logs are still yours to bring.
Policy checks catch a different failure
A simulation tells you what a change blocks. A policy check tells you what a change grants. AWS describes IAM Access Analyzer custom policy checks as a way to compare an updated policy against the existing version and detect new access, and says they can be configured in the development lifecycle to review policies before deployment. That is a statement about logic and supplied inputs, and it makes no claim to know your live traffic.
Run both on a change. The simulation protects availability by showing the legitimate access you would cut. The check protects the boundary by showing access you would add by accident. A pull request that passes only one of them is half reviewed. Teams that run only one of the two checks tend to learn about the other failure in production, from a user who lost access or an auditor who found new access.
A procedure that fits a change window
The sequence below works with any of the tools above and with the vendors on this index. Write the proposed policy as code in a branch. Simulate it against the longest log window the tool allows. Review every changed outcome and name an owner for each one. Run the policy check for new access. Deploy in an audit or log-only mode where the cloud offers one, watch for blocked attempts for a full business cycle, then enforce. Keep the previous policy version ready to restore. Record the result with the change: the simulated outcomes, the reviewer and the date.
- Propose the change as code.
- Simulate against at least 90 days of logs where available.
- Assign an owner to each changed outcome.
- Run a new-access check on the same change.
- Run in audit mode through one business cycle, then enforce.
Vendors add a model of the environment
Act says it tests each recommended change against what it calls a digital twin to show what it would block, and that changes can run in audit mode to confirm that no legitimate access is revoked. Act is read-only, so the customer applies the change through a pull request. Sonrai says policies are built from observed usage, that in-use permissions are exempted first and that the exemption list is shown before deployment.
Ask each vendor what its simulation runs against: cloud configuration, audit-log events, or both. Ask what it cannot cover, such as rarely used or undocumented dependencies. Act and Sonrai Security each have a profile page with the sources read. The two differ in where the last step sits: Sonrai's policies block access once the customer deploys them, and Act opens pull requests the customer merges.
Related pages
Sources
- Google Cloud, Policy Simulator for allow policies, read 10 Nov 2026
- AWS, IAM policy testing with the policy simulator, read 10 Nov 2026
- AWS, IAM Access Analyzer custom policy checks, read 10 Nov 2026
- Google Cloud, Policy Intelligence overview, read 10 Nov 2026
- Act Security, platform page, read 10 Nov 2026
- Sonrai Security, home page, read 10 Nov 2026
Last reviewed 16 Nov 2026.