AI Security Risk Assessment: Protect Cloud AI Platform and Apps You Already Run

Built for the CISO who has been asked by the board what AI the company is using and has to answer honestly that nobody knows. This engagement will eliminate the blind spot around unsanctioned AI usage and reduce the exposure it is already creating.

Who this is for

CISOs who have been asked a question they cannot answer. The trigger is almost always external: a board question, a customer security questionnaire asking about AI usage, a regulator, or an insurer adding an AI clause. The organization has an AI policy or is about to write one, and needs to know the actual position before committing to anything in writing. This assessment produces that position.

What we deliver

• Shadow AI discovery across the estate: which generative AI applications are in use, by how many people, how often, and through what path, covering both sanctioned services and the ones nobody approved.
• Data exposure analysis showing what categories of data are reaching each AI service, with labeled and regulated content called out specifically, and the identities and departments involved.
• An inventory of AI agents, plugins, and connectors already deployed inside your Microsoft estate, including Copilot Studio agents, Power Platform connectors, and third-party applications holding Microsoft Graph permissions, with the permission scope each one actually has.
• A ranked risk register with each finding scored on likelihood and impact, plus remediation options costed at three levels so the response can be matched to the organization's actual risk appetite rather than a maximalist recommendation.

Outcomes our customers see

One assessment across 5,600 seats found 34 generative AI applications in active use against an expected 3, with 12 of them receiving content carrying a confidential label. An agent and permission inventory at a financial services customer surfaced 8 third-party applications holding tenant-wide Microsoft Graph read permissions granted over 500 days earlier and never reviewed. Assessments are delivered in 12 business days, and 11 of the last 12 customers proceeded to remediation within 60 days of receiving the register.

How this compares

The AI risk assessments in this market are mostly questionnaire-based, meaning they document what people say they use. This one measures what is actually happening using Defender for Cloud Apps and Purview telemetry, and the gap between the two is consistently the most useful finding in the report. The agent and Graph permission inventory is the part almost nobody else includes, and it is where the genuinely serious exposure tends to sit.

Architecture and Microsoft alignment

Discovery runs read-only through Microsoft Defender for Cloud Apps for shadow AI application usage, Microsoft Purview Data Security Posture Management for AI for data exposure, Microsoft Entra ID enterprise application and consent records for third-party permission scope, Microsoft Graph for the agent and connector inventory, and Power Platform admin analytics for Copilot Studio agents. Endpoint and network telemetry is used where available to catch usage that does not traverse a Microsoft identity path. No content is read and nothing is changed. Aligned to the Microsoft solution plays Protect Cloud AI Platform and Apps and Data Security.

Plans

Plans, prices, and full scope per plan are on the Plans tab of this listing.

Prerequisites

Global Reader and Security Reader consent, Microsoft 365 E5 or the E5 Security and Compliance add-ons for full discovery fidelity, and Power Platform administrator read access. Where Defender for Cloud Apps is not licensed, discovery fidelity drops materially and that limitation is stated in the report rather than worked around silently.

Limitations

This is a read-only assessment; nothing is blocked, revoked, or remediated, and remediation is quoted separately. AI usage on personal devices outside all managed telemetry is not discoverable and the report says so explicitly rather than implying full coverage. The assessment covers AI consumption risk; it does not review the security of AI systems your own engineering teams are building, which is a separate engagement.

How to buy

Buy through the Azure portal, using Get it in Azure portal on this listing, so the purchase is billed through your existing Microsoft agreement. Private offers on request.

Next step

Get it now in the Azure portal, or request a private offer if the scope or the price needs adjusting first.