Agent Impersonation

Expanded Definition

Agent impersonation is more than simple spoofing. In NHI and agentic AI environments, it occurs when a malicious actor convinces a user, system, or downstream agent that a fraudulent identity is the trusted autonomous entity authorized to issue instructions, approve actions, or request credentials. The distinction matters because the target is often not a human reviewer but another machine, orchestration layer, or delegated workflow that trusts identity signals too readily.

Definitions vary across vendors on whether the term includes only fully forged agent identities or also stolen credentials, session hijacking, and prompt-level deception. In practice, the security failure is the same: trust is granted to an agent that has not been properly authenticated or bound to the expected permissions. Guidance in the OWASP Agentic AI Top 10 and the NIST AI Risk Management Framework both point toward stronger identity proofing, authorization boundaries, and traceable agent actions. The most common misapplication is treating agent identity as a naming convention instead of a cryptographically enforced trust relationship, which occurs when orchestration systems accept self-asserted agent claims without verifying signatures or provenance.

Examples and Use Cases

Implementing agent identity rigorously often introduces latency and operational overhead, requiring organisations to weigh faster delegation against stronger verification and tighter approval paths.

• A fraudulent support agent sends a tool-use request that looks like a legitimate escalation, but the receiver validates only the displayed agent name and not the signed identity.
• A compromised automation service replays prior approvals to impersonate a trusted planner agent and trigger privileged workflows.
• An attacker clones an agent endpoint in a multi-agent system so a downstream agent accepts fake policy updates and exposes secrets.
• A workflow platform trusts an LLM agent’s self-declared role, allowing it to request actions outside the intended permission boundary.
• Investigations into incidents such as the CoPhish OAuth Token Theft via Copilot Studio and Gemini AI Breach — Google Calendar Prompt Injection show how identity confusion and untrusted instructions can cascade across agentic systems.

For teams building controls, the relevant standard references are the OWASP Top 10 for Agentic Applications 2026 and CSA MAESTRO agentic AI threat modeling framework, which both help teams reason about trust, delegation, and control boundaries.

Why It Matters in NHI Security

Agent impersonation becomes a governance problem when organisations cannot reliably tell which autonomous entity approved a request, touched a secret, or invoked a tool. That ambiguity breaks auditability, weakens least privilege, and allows one compromised identity to masquerade as another across workflows, APIs, and approval chains. In NHI-heavy environments, the blast radius is rarely limited to a single system; it often spreads through service accounts, tokens, and delegated trust paths that were assumed to be authentic.

NHI Management Group reports that 80% of identity breaches involved compromised non-human identities such as service accounts and API keys, which is directly relevant because impersonation often begins with the theft or misuse of those same credentials. See the Ultimate Guide to NHIs for the broader governance context and the OWASP NHI Top 10 for the control implications of identity abuse. Once impersonation has occurred, responders typically discover the issue only after an unexpected approval, data access, or destructive action, at which point agent identity verification becomes operationally unavoidable to address.

Related resources from NHI Mgmt Group

• AI Agent Authentication
• Agent Certificate Impersonation
• What is agent communication poisoning?
• What is a shadow agent and why is it more dangerous than a typical shadow NHI?