A CIEM requirements checklist: 15 requirements, each with a test you can run in a trial and the criterion it maps to
Facts read 16 Nov 2026. Page reviewed 16 Nov 2026.
A CIEM buyer's checklist of 15 requirements, each with the reason it matters, a test to run during a trial and the one of the seven index criteria it maps to. The criteria and weights, fixed before scoring, are removal 25, safety 20, class of risk 15, drift 15, coverage 10, native controls 10 and audit 5.
Fifteen requirements and the test for each
Run each test on a non-production account with realistic roles. The last column uses the seven criteria of the method, so a result can be carried into the scoring sheet in the CIEM RFP template.
| Requirement | Why | How to test in a trial | Criterion |
|---|---|---|---|
| Ends in a removal, not only a finding | The product that only reports leaves the work to your team | Pick 10 unused grants from the product's own report and ask it to remove them; time the whole path | Removes access, not only reports it |
| Names what it will change before it changes it | Removing a permission in production can break a workload | Request a change and ask for the list of identities and actions affected before approval | Safe to enforce |
| Simulates against your real policies | A simulation on sample data proves nothing about your account | Submit a change you know would break something; confirm the tool flags it | Safe to enforce |
| Rollback or restore path | A block with no way back pushes engineers to bypass the control | Remove access, then restore it; record the minutes taken and the steps | Safe to enforce |
| One change covers a class of paths | Permission-by-permission clean-up never finishes | Ask for a single boundary or guardrail that blocks a risk across all accounts; count the paths it removes | One change, a class of risk |
| Applies at organization level | Account-by-account policy leaves new accounts open | Create a new test account and see whether the control applies on day one | One change, a class of risk |
| Checks changes before production | Access removed today can be granted again tomorrow | Submit an infrastructure-as-code change that adds a broad grant; confirm it is flagged | Stays fixed |
| Exceptions handled in your own tools | Exceptions in a separate queue get lost | Request an exception from your ticketing or chat tool; confirm it reaches the product | Stays fixed |
| Proof that a fix held | An audit asks for evidence over months, not a screenshot | Ask for a report of one fixed finding and its state 30 days later | Stays fixed |
| Covers identity, network and AI workloads together | Excess is often reachable only through a network path | Pick a workload with an AI service role; confirm it appears with its network exposure | Identity, network and AI access together |
| Machine identities included | Service accounts and workload roles outnumber people | Ask for the count of non-human identities in a sample account and check it against your own list | Identity, network and AI access together |
| Uses the cloud's own policy types | Nothing new in the request path means nothing new to fail | Ask which native mechanism enforces each change (service control policy, role, deny policy) and read one of them in your console | Uses the cloud's own controls |
| No proxy or broker in the request path | A broker is an outage risk | Disable the product's connection during the trial; confirm access still works | Uses the cloud's own controls |
| Mapped controls and continuous evidence | Audits ask for the control, the date and the owner | Export evidence for one named framework and check it against your auditor's list | Evidence for auditors |
| Entitlement count with a stated unit | Counts differ by unit (policy, role, assignment) | Ask for the unit and the total on one sample account; compare with your own count | Identity, network and AI access together |
How the requirements map to the weights
Removal carries 25 of 100, safe enforcement 20, one change for a class of risk 15, staying fixed 15, identity, network and AI coverage 10, native controls 10 and evidence for auditors 5. The weights were fixed on 10 Nov 2026 before any vendor was scored. Six of the 15 requirements above map to the top three criteria, which is where most of the weight lies. Per-criterion scores for all 14 platforms are in the criteria comparison.
If your own priorities differ, change the weights before you open any vendor response, not after. The situation pages such as AWS-heavy organizations show how published weights change the order.
What a trial cannot show
A two-week trial shows whether a product finds excess and whether its removal works on the accounts you gave it. It rarely shows behavior at several hundred accounts, how often the vendor's connector breaks after a cloud API change, or the support response at month nine. Ask for two references of similar size and ask each what broke.
Ask for pricing early. The CIEM pricing page lists which vendors publish a price and which ask for a quote, as read on 16 Nov 2026. Where the answer is a quote, ask for the unit before the number.
Related pages
Sources
- Wiz Academy, how to choose a CIEM vendor, read 16 Nov 2026
- Microsoft Learn, CIEM in Defender for Cloud, read 16 Nov 2026
- AWS documentation, IAM Access Analyzer, read 16 Nov 2026
- Google Cloud, IAM role recommendations overview, read 16 Nov 2026
Last reviewed 16 Nov 2026.