AI Agent Identities Are Redrawing Enterprise IAM Budgets

Enterprise AI agent projects are multiplying faster than identity governance programmes can accommodate them, and the funding patterns required to secure those agents differ materially from conventional identity and access management (IAM) spending, according to new research published by Omdia.

The core finding is straightforward. AI agents operate as non-human identities: they authenticate, they hold permissions, and they act on behalf of — or independently of — human users. But existing IAM budget lines were built around provisioning and de-provisioning people. That structural mismatch is now producing measurable gaps.

Omdia's analysts note that organisations running AI agent workloads are encountering credential management questions that sit outside the scope of traditional human identity governance. An AI agent might authenticate to a database, call an external API, and write to a cloud storage bucket within a single workflow (each action leaving a permissions footprint that conventional role-based access control was not designed to audit at that velocity). The policy question is not merely technical; it concerns who owns the governance obligation when an agent acts.

And that ownership question is unsettled. Neither the security team nor the application development team has consistently claimed it, said analysts at Omdia in the research note. Budget authority tends to follow ownership, which means AI agent identity security is frequently funded from discretionary or project-level allocations rather than from the standing IAM programme.

The Cybersecurity and Infrastructure Security Agency (CISA) has flagged non-human identity risk in its guidance on secure-by-design principles, though CISA has not yet issued a regulation specific to AI agent credential management. The National Institute of Standards and Technology (NIST) addresses machine identity in SP 800-63C, the federation and assertion standard, but that document predates agentic AI architectures and does not cover the lifecycle management questions those architectures raise.

So the regulatory gap is real. Organisations subject to the Securities and Exchange Commission's (SEC) cybersecurity disclosure rules — effective for large accelerated filers since December 15, 2023 — must disclose material cybersecurity risks and incidents, and a misconfigured AI agent identity that enables unauthorised data access would likely meet that materiality threshold. Whether internal audit functions are reviewing AI agent permissions with the same rigour applied to human privileged accounts is, at present, inconsistent across industries.

The practical implication for compliance officers is procedural. If AI agent identities are not enumerated in the organisation's identity inventory, they cannot be scoped into access reviews, cannot be included in privileged access management controls, and will not appear in the evidence packages submitted to auditors under frameworks such as SOC 2 or ISO 27001.

But the budget dynamic identified by Omdia may be the more immediate pressure. Project teams funding AI agent deployments are absorbing identity security costs that IAM programme owners do not see, which means aggregate spend on identity governance is likely underreported at the enterprise level.

The Omdia research does not carry a formal comment period. Organisations assessing their own exposure should consult NIST SP 800-63C and the CISA secure-by-design guidance as interim reference points until agency-specific standards emerge.