Is default abuse monitoring for Azure OpenAI (Foundry Models sold by Azure) covered by the HIPAA BAA?

Question

Is default abuse monitoring for Azure OpenAI (Foundry Models sold by Azure) covered by the HIPAA BAA?

Alden Haight 0

We are a HIPAA covered entity / business associate with the Microsoft BAA executed via the DPA / Product Terms. We are planning to process protected health information (PHI) through Azure OpenAI models ("Foundry Models sold by Azure") using Data Zone Standard deployments in a US region, private endpoints, and Entra ID-only authentication. We will be operating with default abuse monitoring — we are not a managed customer and have not been granted the modified abuse monitoring exemption.

We have reviewed the data privacy documentation and the abuse monitoring documentation, and have four questions we could not resolve from them:

BAA scope. Is the abuse monitoring process — including the temporary data store for flagged prompts/completions and any human review — performed by Microsoft as part of providing the Online Service, within the scope of Microsoft's obligations under the HIPAA BAA (i.e., in its capacity as a business associate)? Does BAA coverage depend in any way on obtaining the modified abuse monitoring exemption?
Retention. The documentation previously stated abuse-monitoring data was stored "up to thirty (30) days"; that language was removed around the November 18, 2024 update that introduced AI-based review. What is the current retention commitment for flagged prompts/completions held for human review?
Reviewer location. For a resource deployed in the United States geography, is access to the abuse-monitoring data store (and any human review) restricted to US-located personnel and systems? The documentation commits to EEA-located reviewers for EEA deployments — is there an equivalent US commitment?
Where the BAA enumerates this service. Where is "Foundry Models sold by Azure" / Azure OpenAI listed as in-scope for the HIPAA BAA (the services appendix in the BAA at aka.ms/BAA, or the Azure Compliance Offerings appendices on the Service Trust Portal)? Does default abuse monitoring alter that scope in any way?

Documentation-grounded answers we can cite in a compliance risk analysis would be greatly appreciated.

Manish Deshpande 7,770 Reputation points Microsoft External Staff Moderator

2026-07-08T22:23:44.7766667+00:00
Hello @Alden Haight ,

One framing note up front: everything Microsoft does to deliver the service to you including abuse monitoring is processing of your Customer Data under the Products and Services Data Protection Addendum (DPA), which is where the HIPAA BAA is established.

So the right lens isn't "is abuse monitoring separately covered," it's "abuse monitoring is part of the covered Online Service."

1) BAA scope, and dependence on the modified exemption

Per the documentation, the abuse-monitoring flow the logically-isolated store of flagged prompts/completions and any human review is performed by Microsoft as part of providing the service, and all such processing of your prompts/completions is governed by the DPA (see the Data, privacy, and security doc, which states the DPA governs data processing for these models). Because the HIPAA BAA is executed through the DPA for covered entities/business associates, that processing sits inside the same contractual framework it does not require the modified abuse-monitoring exemption. The exemption changes how your data is handled (it turns off the temporary storage and human-review step); it does not determine whether the BAA framework applies. Running default monitoring, as you plan, stays within that framework.

Two caveats for your file: (a) treat the binding scope determination as a contractual question — cite the DPA/Product Terms and, if you want it stated unambiguously for HIPAA, get written confirmation from your Microsoft account/licensing contact rather than relying on this thread; (b) for assurance on the human-review controls themselves, the docs state review is limited to authorized Microsoft employees, accessed via Secure Access Workstations (SAWs) with Just-In-Time (JIT) manager approval, using point-wise request-ID queries.

2) Current retention commitment

You're right that the earlier "up to thirty (30) days" language was removed around the November 18, 2024 update that introduced AI-based review. The current Data, privacy, and security document does not state a fixed day-count; it describes the flagged data as held in the customer-isolated, in-geography abuse-monitoring store for the review process. For a compliance file I'd recommend: cite the current retention clause directly from that legal document as of your assessment date (it's the versioned, binding source — not the concept page), capture a dated copy for your evidence pack, and if you need the exact retention window for human-review-held data stated explicitly, obtain it in writing via your account team. That page also documents how to verify abuse-monitoring storage state on your resource (the ContentLogging capability in the resource JSON view / az cognitiveservices account show), which is useful audit evidence.

3) US reviewer-location commitment

To be straight with you: the documentation makes an explicit commitment that for resources deployed in the EEA, human review is performed by authorized Microsoft employees located in the EEA. There is no published, symmetric statement guaranteeing US-only personnel for US deployments. What is committed for a US (DataZone, US) deployment is data-at-rest residency: the abuse-monitoring data store for Global/DataZone deployments is kept in the customer-designated geography (US stays US). That's a residency commitment about where the data sits — it is not the same as a commitment about where reviewer personnel are located. I'd flag the US reviewer-location point as an open item to confirm in writing with Microsoft for your risk analysis, rather than inferring parity with the EEA language.

4) Where the service is enumerated as in-scope, and whether monitoring changes it

The BAA is established through the DPA — there's no separate signature for this service; execution of your volume licensing agreement includes execution of the HIPAA BAA. The authoritative in-scope enumeration lives in the sources you'd cite in an audit:

the HIPAA in-scope services listing via the Service Trust Portal / Microsoft Learn compliance offerings, and

the DPA / Product Terms referenced from aka.ms/BAA (the contractual list).

Azure OpenAI now surfaced as "Foundry Models sold by Azure" is covered as part of the in-scope Azure AI services under that framework. Default abuse monitoring does not alter that scope: it's a feature of the covered service, not a separate service outside the BAA. Because these lists are versioned, pull the in-scope list from the STP/DPA as of your assessment date for your binder.

A practical note for the risk analysis: The clean path to contractual/written confirmation is your Microsoft account team / privacy contact, the Service Trust Portal documents, or an Azure support case referencing your subscription. And if you ultimately decide you need the temporary storage and human review turned off (rather than just confirmation that default monitoring is BAA-covered), that's the modified abuse monitoring path it has its own Limited Access eligibility and is separate from, and not required for, BAA coverage.

Links:
https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy?tabs=azure-portal
https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/abuse-monitoring

https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/limited-access

https://www.microsoft.com/licensing/docs/view/Microsoft-Products-and-Services-Data-Protection-Addendum-DPA

https://servicetrust.microsoft.com/

Thanks,
Manish.
Manish Deshpande 7,770 Reputation points Microsoft External Staff Moderator

2026-07-16T01:08:14.5+00:00

Hello @Alden Haight

I just wanted to kindly follow up to check if you had a chance to review my earlier response. I hope the information shared was helpful. Please feel free to reach out if you have any additional questions or need further clarification.

1 answer

Your answer

Manish Deshpande 7,770 Reputation points Microsoft External Staff Moderator

2026-07-16T01:08:14.5+00:00

Hello @Alden Haight

I just wanted to kindly follow up to check if you had a chance to review my earlier response. I hope the information shared was helpful. Please feel free to reach out if you have any additional questions or need further clarification.

Answer 1

Hi! Microsoft documents that flagged prompts and completions may be placed in a geographically separated abuse-monitoring store and accessed by authorized Microsoft reviewers when necessary. Modified abuse monitoring removes that storage and human-review process, but it is available only to eligible customers. The public documentation does not provide enough contractual detail to conclusively answer your BAA-scope and US-reviewer questions. Obtain written confirmation from Microsoft compliance or your account team before sending PHI. Relevant references are Azure Direct Models data privacy and Azure HIPAA compliance.