Use Microsoft Global Secure Access to inspect traffic patterns that reveal AI agent use, then map those sessions back to users, applications and destinations. Look for unusual access to AI services, automation-like request bursts, and connections from identities that should not be using them. Correlate network signals with identity logs to confirm what is actually in use.
How Microsoft Global Secure Access helps you spot AI agent activity
Global Secure Access is useful because it gives you network-level visibility into where traffic is going, which identities are making the calls, and whether the pattern looks like a person using an app or an automated agent. For discovery, the key is not a single signature, but a combination of destinations, cadence, and identity context that can be validated against logs.
That makes it a practical starting point for AI agent discovery in environments where agents are not centrally registered. A session that repeatedly reaches AI services, model endpoints, or agent-hosting platforms may be worth closer review when the calling identity, source device, or application context does not match the expected business use.
Microsoft documents the product as a way to secure and monitor access paths, and that is exactly why it can help with discovery, because the traffic itself becomes evidence you can compare with identity records and sanctioned application inventories. The result is a discovery process based on observed behaviour, not assumptions about what teams say they deployed.
What traffic patterns usually indicate an AI agent
The most useful signals are behavioural. Look for repeated requests to AI services, bursty or machine-like request timing, and connections that continue across long work periods without a matching human interaction pattern. If the same identity generates frequent calls to model APIs, copilots, or automation backends, that may indicate an agent, an integration, or a human-operated script.
It is also important to distinguish normal enterprise AI use from unmanaged agent use. A sanctioned application may generate similar traffic, but it should map to a known owner, approved destination, and approved purpose. Unfamiliar endpoints, unusual user agents, or destinations that appear only from a small set of identities are stronger discovery clues than volume alone.
Shadow AI and AI Agent Discovery Guide is the most direct internal reference for building that inventory logic from OAuth grants, API keys, cloud, endpoint and network signals. For practitioners, the real task is to separate sanctioned AI usage from the shadow AI and unmanaged agent patterns that create governance blind spots.
How to confirm what is actually in use
Discovery becomes trustworthy when you correlate network telemetry with identity and application evidence. Use Global Secure Access to identify the session, then match it to sign-in logs, app registrations, consent grants, service principals, or other records that show who or what initiated the traffic. If the network trail exists but no corresponding approved asset exists, treat it as an inventory gap until proven otherwise.
Correlation should also tell you whether the activity is tied to a user, a service, or an automation path. Some AI agents ride on user tokens, some use app credentials, and some are embedded in SaaS workflows. Those differences matter because the remediation path changes: a human-account issue calls for user review and privilege checks, while an app-backed agent may require consent review, credential rotation, or control of the service principal.
AI Agent Observability, Audit and Incident Response Guide is useful here because it reinforces the need for attribution, logging, and tested response when agent behaviour must be explained after the fact. In practice, discovery is only complete when you can answer who made the call, which identity was used, and whether the traffic matches an approved use case.
Risk and Threat Considerations
Untracked AI agents create exposure because they may use sanctioned identities, approved network paths, or legitimate AI services in ways the business never intended. The risk is not just accidental sprawl, it is also loss of control over data movement, third-party access, and autonomous actions that can scale faster than manual review.
Failure mechanism: An agent can blend into normal cloud or browser traffic while using overbroad permissions, inherited tokens, or hidden automation paths, which makes it harder to distinguish legitimate use from unauthorized or excessive use.
Impact: Teams may miss unauthorized data access, unexpected API consumption, or a broader compromise that starts with what looks like ordinary AI traffic but actually represents unmanaged automation or abused credentials.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Non-Human Identity Top 10 and OWASP Agentic AI Top 10 address the attack and risk surface, while NIST SP 800-53 Rev 5 sets the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Non-Human Identity Top 10 | NHI-03 — Vulnerable Third-Party NHI | AI agents discovered through network traffic may expose third-party identity and access paths. |
| NHI-05 — Overprivileged NHI | Discovery helps identify agents using excessive permissions or broader-than-expected access. | |
| NHI-10 — Human Use of NHI | Network discovery can reveal when humans are driving agent-like access patterns or borrowed identities. | |
| Recommendation — Review third-party agent integrations and constrain their access paths. Reduce agent permissions to the minimum needed for each approved workflow. Separate human activity from agent access paths and prohibit credential sharing. | ||
| OWASP Agentic AI Top 10 | ASI03 — Identity & Privilege Abuse | AI agent discovery focuses on identities and privileges used for automated actions. |
| Recommendation — Bind each agent to a distinct identity and limit its privileges per action. | ||
| NIST SP 800-53 Rev 5 | AU-6 — Audit Review, Analysis, and Reporting | Discovery depends on correlating network and identity logs to identify agent activity. |
| IA-5 — Authenticator Management | AI agent discovery often leads to managing the credentials behind the observed traffic. | |
| Recommendation — Correlate logs to detect unusual AI-service access and trace it to a source identity. Rotate or revoke credentials when discovered agent traffic is tied to unmanaged access. | ||
Practitioner Guidance
What to verify: Start with the destination list, then confirm whether each AI-related destination has a business owner, approved identity, and documented purpose. If you cannot tie the session back to an approved application or workflow, treat it as an unmanaged AI use case until the owner proves otherwise.
Decision rule: If the traffic is high-confidence AI activity but the identity is unexpected, prioritise identity and consent review before trying to label the traffic as benign. If the pattern is repetitive and machine-like, assume automation until evidence shows a human-driven workflow.
Practitioner takeaway: The best discovery outcome is not just finding AI traffic, it is being able to prove whether that traffic belongs to an approved agent, an embedded integration, or shadow automation that needs governance.
Related resources from NHI Mgmt Group
Deepen Your Knowledge
Free weekly newsletter
Subscribe to the NHI & AI Identity Journal
The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.
Bonus 33% off our NHI Course when you subscribe.
Reviewed and updated by the NHIMG editorial team on October 20, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org