# Shadow AI detection evidence for auditors

URL: https://shadowaidetectiontools.com/guides/shadow-ai-detection-evidence-for-auditors/
Updated: 2026-11-16

Auditors ask for a dated inventory of AI tools, the sources it came from, who approved each tool, and records showing the control ran. Keep detection logs for the period your audit scope requires, and note that platform defaults differ: Microsoft Audit (Standard) keeps records 180 days.

## What evidence do auditors ask for?

Auditors ask for five things: the inventory with dates, the discovery method and sources, the approval record per tool, proof the control operated over the period, and exceptions with sign-off. A screenshot of a dashboard shows one day; a dated export shows the period.

Pick tools that export. Harmonic's documentation describes CSV export of the per-app risk profile, Zscaler documents an Excel export for its MCP inventory, and for several products an inventory export was not found on 16 Nov 2026 in the vendor files.

Sources: [Harmonic Security: shadow AI detection](https://www.harmonic.security/solutions/shadow-ai-detection); [Zscaler: AI assets, MCP servers](https://help.zscaler.com/ai-asset-mgmt/about-ai-assets-mcp-servers); [Harmonic Security source (harmonic-security-36)](https://docs.harmonicsecurity.app/llms.txt).

## How long should detection logs be kept?

Keep them for the period set by your audit scope or regulation, and confirm what the platform keeps by default. Microsoft Purview Audit (Standard) retains audit records for 180 days, and the page notes that longer retention requires other options, such as exporting through the Office 365 Management Activity API to a SIEM.

Cloud discovery and endpoint tools have their own retention, so ask each vendor. The answer belongs in the evidence file next to the export schedule.

Sources: [Auditing solutions in Microsoft Purview](https://learn.microsoft.com/en-us/purview/audit-solutions-overview).

## How do you show the control actually ran?

Show scheduled runs, alerts raised, and dispositions closed. Cloud discovery policies that alert on new apps leave a record of alerts, and identity logs show consent grants and revocations by date. Entra ID audit logs capture changes to applications, groups, users and licenses.

Pair each alert with the ticket that closed it. Several products route to a ticket: Push Security documents Jira tickets through a webhook, and Cyberhaven documents ServiceNow and SOAR integrations.

Sources: [Create cloud discovery policies, Microsoft Defender for Cloud Apps](https://learn.microsoft.com/en-us/defender-cloud-apps/cloud-discovery-policies); [Audit logs in Microsoft Entra ID](https://learn.microsoft.com/en-us/entra/identity/monitoring-health/concept-audit-logs); [Push Security: shadow AI, how to discover, govern and secure AI apps](https://pushsecurity.com/blog/shadow-ai-how-to-discover-govern-and-secure-ai-apps); [Cyberhaven: AI security](https://www.cyberhaven.com/product/ai-security); [Push Security help: sending Push data to a SIEM or ticketing system](https://pushsecurity.com/help/can-i-send-push-data-to-other-systems-like-a-siem-or-ticketing-system); [Cyberhaven integrations](https://www.cyberhaven.com/product/integrations).

## What about risk assessments of each tool?

Keep the assessment on file per approved tool: data processed, training and retention terms, hosting, certifications and the decision date. Vendors provide starting data. Microsoft's catalog scores more than 31,000 apps on 90-plus risk factors, Nudge Security offers over 200,000 SaaS vendor security profiles, and Zscaler documents a risk index with hosting, posture and retention.

A vendor score is input, not the assessment. The reviewer's decision and date are the evidence.

Sources: [Cloud app discovery overview, Microsoft Defender for Cloud Apps](https://learn.microsoft.com/en-us/defender-cloud-apps/set-up-cloud-discovery); [Nudge Security](https://www.nudgesecurity.com/); [Zscaler: public AI inventory, GenAI](https://help.zscaler.com/ai-asset-mgmt/about-public-ai-inventory-gen-ai).

## What does a minimal evidence pack look like?

A minimal pack holds the policy, the scope statement, the dated inventory exports, the source list, the approval records, alert and ticket samples, and the exception register. Store it where the auditor can read it without asking the tool owner.

Assemble it quarterly so no audit starts from zero.

## What gaps do auditors find most often?

The common gaps are an inventory with no dates, no stated scope, no owner on entries, and exceptions without sign-off. A second gap is relying on one source: an auditor can ask which devices and tenants were covered, and the answer must be a number.

Fix them by adding the coverage line to the inventory header, storing the policy version with the export, and keeping exceptions in a register with an expiry date. None of these needs a new tool.

If a tool is added mid-period, record the date it went live. An auditor reading a count that rises sharply will look for the reason, and a dated source list explains it before the question is asked.

Sources: [Auditing solutions in Microsoft Purview](https://learn.microsoft.com/en-us/purview/audit-solutions-overview).

## Sources

- [Harmonic Security: shadow AI detection](https://www.harmonic.security/solutions/shadow-ai-detection) (harmonic.security), read 2026-11-16
- [Zscaler: AI assets, MCP servers](https://help.zscaler.com/ai-asset-mgmt/about-ai-assets-mcp-servers) (help.zscaler.com), read 2026-11-16
- [Auditing solutions in Microsoft Purview](https://learn.microsoft.com/en-us/purview/audit-solutions-overview) (learn.microsoft.com), read 2026-11-16
- [Create cloud discovery policies, Microsoft Defender for Cloud Apps](https://learn.microsoft.com/en-us/defender-cloud-apps/cloud-discovery-policies) (learn.microsoft.com), read 2026-11-16
- [Audit logs in Microsoft Entra ID](https://learn.microsoft.com/en-us/entra/identity/monitoring-health/concept-audit-logs) (learn.microsoft.com), read 2026-11-16
- [Push Security: shadow AI, how to discover, govern and secure AI apps](https://pushsecurity.com/blog/shadow-ai-how-to-discover-govern-and-secure-ai-apps) (pushsecurity.com), read 2026-11-16
- [Cyberhaven: AI security](https://www.cyberhaven.com/product/ai-security) (cyberhaven.com), read 2026-11-16
- [Cloud app discovery overview, Microsoft Defender for Cloud Apps](https://learn.microsoft.com/en-us/defender-cloud-apps/set-up-cloud-discovery) (learn.microsoft.com), read 2026-11-16
- [Nudge Security](https://www.nudgesecurity.com/) (nudgesecurity.com), read 2026-11-16
- [Zscaler: public AI inventory, GenAI](https://help.zscaler.com/ai-asset-mgmt/about-public-ai-inventory-gen-ai) (help.zscaler.com), read 2026-11-16

## Related pages

- [What a Shadow AI Inventory Should Contain](/guides/what-a-shadow-ai-inventory-should-contain/)
- [Shadow AI Metrics for the Board](/guides/shadow-ai-metrics-for-the-board/)
- [AI asset register (glossary)](/glossary/ai-asset-register/)
- [All guides](/guides/)
- [Reading Proxy and DNS Logs for AI Apps](/guides/reading-proxy-and-dns-logs-for-ai-apps/)
- [Shadow AI Detection in Microsoft 365 and Google Workspace](/guides/shadow-ai-detection-in-microsoft-365-and-google-workspace/)
- [The ranking](/ranking/)
- [Glossary](/glossary/)
