# Can only approved models and AI services be used in your accounts, enforced by policy?

Facts read 16 Nov 2026. Page reviewed 16 Nov 2026.

Question 25 of the assessment asks: Can only approved models and AI services be used in your accounts, enforced by policy? Unapproved models and endpoints send your data to places you have not assessed. It comes from Microsoft cloud security benchmark AI-1 and carries weight 2 in the ai workloads domain.

## Why it matters

Unapproved models and endpoints send your data to places you have not assessed.

## Answers and what each is worth

| Answer | Value |
| --- | --- |
| Yes, enforced by policy | 1 |
| By guidance only | 0.5 |
| No or not known | 0 |

## How to check it in each cloud

- **AWS:** Check SCPs or IAM policies that limit bedrock:InvokeModel to approved model ARNs, and that other AI services are denied where not needed.
- **Azure:** Check Azure Policy assignments that restrict model deployments and AI service resource types to the approved set.
- **Google Cloud:** Check the Vertex AI model garden and org policy constraint for restricting allowed models and services.

## The fix

Enforce an allow list of models and AI services with organization policies. Effort: low.

## Where it comes from

- Microsoft cloud security benchmark AI-1: [Ensure use of approved models](https://learn.microsoft.com/en-us/security/benchmark/azure/mcsb-v2-artificial-intelligence-security)
- NIST SP 800-53 Rev. 5 CM-2: [source](https://csrc.nist.gov/pubs/sp/800/53/r5/upd1/final)

## Related pages

- [Question 26: AI workloads, ai data reach](/questions/ai-data-reach/)
- [Question 27: AI workloads, human approval](/questions/human-approval/)
- [Question 28: AI workloads, ai activity logging](/questions/ai-activity-logging/)

## Sources

- Microsoft cloud security benchmark AI-1: https://learn.microsoft.com/en-us/security/benchmark/azure/mcsb-v2-artificial-intelligence-security (read 16 Nov 2026)
- NIST SP 800-53 Rev. 5 CM-2: https://csrc.nist.gov/pubs/sp/800/53/r5/upd1/final (read 16 Nov 2026)

Last reviewed 16 Nov 2026. Canonical: https://cloudsecurityriskassessment.com/questions/approved-models-only/
