An operating model, not a one-time project
AI tools change weekly. The model that works has four parts: a decision for every app, controls that enforce the decision, logs that prove it, and a request path so staff do not route around you.

Sanction, monitor or block: the decision flow
The four parts of the model
Catalog
An approved-AI list with the plan, owner, allowed data types and a one-line rule for each tool. Published where staff look.
Controls
Unsanctioned apps blocked on managed devices, SSO and MFA on approved ones, DLP at the upload boundary, extensions by policy.
Logging and alerts
Prompt and activity logs where the platform offers them, Entra sign-in logs for AI apps, alerts on new AI apps and on uploads of labeled data.
Request path
A form and a two-week review promise. Staff ask instead of hiding, and every request becomes an inventory entry.
What to log, where, and how long
| Signal | Where it lives | Keep for | Alert when |
|---|---|---|---|
| New generative-AI app seen | Defender for Cloud Apps discovery | 12 months | Any new app with risk score below 6 |
| Sign-ins to AI apps | Entra ID sign-in logs | 12 months | Sign-in from unmanaged device or new country |
| Uploads of labeled files to AI apps | Purview DLP, endpoint DLP | 12 months | Any Confidential or Highly Confidential label |
| Copilot prompts and responses | Purview audit log, DSPM for AI | As retention policy | Prompts containing sensitive info types |
| Agent tool calls and approvals | Agent platform logs, Entra Agent ID logs | 12 months | Action outside the agent's allowed tool list |
| Admin changes to AI settings | Microsoft 365 unified audit log | 12 months | Any change outside a change ticket |

Why this matters in 2026
A page from the sample report
Every finding in the real report has this shape: domain, finding, severity, evidence, fix. See the full sample report.
Guided by Ali Hassani, CISO
Certified CISO (C|CISO) and CISSP with 25+ years in IT, cybersecurity, compliance and infrastructure, and dozens of Microsoft and Cisco certifications. Every AI Security Audit is read-only, evidence-based, and written so an owner and an administrator can both act on it.
Questions business leaders ask
Can we monitor prompts without reading employees' private messages?
Monitor at the boundary: which app, which account, which label of file. Prompt content is logged only in enterprise AI platforms the company owns, under the acceptable-use policy staff signed.
Do we need new tools?
Most Microsoft 365 Business Premium and E3/E5 tenants already have Defender for Cloud Apps discovery, Entra logs and Purview DLP. The audit tells you which licenses you already pay for and have not switched on.
Who owns the catalog?
The IT manager owns it; a business owner approves each entry; security reviews the risk. The roles page has the full chart.
Implementation partner
OC Security Audit stays independent and read-only. If you want hands-on administration, IT Perfection can implement the fixes: DLP rules, Conditional Access, app blocking, agent identities, and ongoing Microsoft 365 support.
Need the operating model set up, not just described?
The AI Security Audit documents the decision for every AI app in use and gives IT a logging and alerting plan it can switch on.
Read-only audit. No changes to your systems. Irvine, Orange County and Southern California on-site; remote nationwide.