Notifications
Clear all
Topic starter
14/09/2026 10:09 am
TL;DR: An agent can have valid credentials and still execute an impersonated request, because authorization proves only that the software may act, not that the human behind the request is genuine, according to Trusona. The governance gap is that verification and delegation are being treated as the same problem when they are not.
NHIMG editorial — based on content published by Trusona: AI agents need human identity verification, not just credentials
By the numbers:
- Mandiant found voice phishing involved in 23% of cloud intrusions and in 11% of intrusions overall.
- 80% of organisations report their AI agents have already performed actions beyond their intended scope.
Questions worth separating out
Q: How should security teams handle delegated access when AI agents act on behalf of customers?
A: Security teams should treat delegated access as a separate governance layer, not as a normal login session.
Q: Why are AI agent credentials not enough to prevent impersonation?
A: Because credentials establish authorization for the agent, not authenticity of the underlying request.
Q: What breaks when humans verify AI output but do not own the workflow?
A: Responsibility becomes ambiguous.
Practitioner guidance
- Define which agent actions require external human verification Map account recovery, MFA reset, enrollment changes, and other high-impact actions to a verification step that checks the human behind the request against an authority the requester does not control.
- Separate NHI credential controls from request authenticity controls Keep token rotation, scoping, and ownership reviews in the NHI programme, but add a distinct control for proving request legitimacy before the agent executes sensitive changes.
- Require API-callable verification for agent workflows Use a verification method the agent can initiate, poll, and branch on, rather than a human-only page that breaks automation or invites bypass.
What's in the full article
Trusona's full blog covers the operational detail this post intentionally leaves for the source:
- Runnable agent-skill setup for ATO Protect, including the shell-based workflow and environment variable configuration.
- The exact verification sequence for creating, polling, and resolving a human identity check inside an agent workflow.
- Implementation details for running the check against external authority records rather than requester-controlled data.
- The distinction between the software-agent use case and the separate human help desk Agent Verify workflow.
👉 Read Trusona's blog on verifying the human behind AI agent actions →
AI agent identity verification: what controls are teams missing?
Explore further
14/09/2026 10:10 am
Identity verification and NHI credential management are not interchangeable controls: This article exposes a common governance mistake, which is treating an agent's ability to act as proof that the underlying request is legitimate. NHI hygiene can tell you who the software is, but it cannot tell you who asked the software to act. Practitioners should stop collapsing these controls into one workflow.
A few things that frame the scale:
- Only 52% of companies can track and audit the data their AI agents access, leaving 48% with a complete blind spot for compliance and breach investigation, according to AI Agents: The New Attack Surface report.
- 80% of organisations report their AI agents have already performed actions beyond their intended scope, including accessing unauthorised systems, sharing sensitive data, and revealing access credentials.
A question worth separating out:
Q: Should organisations use one control for both NHI governance and human request verification?
A: No. NHI governance should manage the agent's token, scope, and ownership, while request verification should prove the human behind the action. Combining them hides the failure mode and gives a false sense of security, especially in delegated access and identity recovery flows.
👉 Read our full editorial: AI agents need human identity verification, not just credentials
