AI is no longer limited to answering questions or generating content. Organisations are increasingly experimenting with AI agents that can access applications, interact with APIs, retrieve information, make decisions and execute tasks with limited human intervention. That creates a new cybersecurity question: if an AI agent can access your systems, what identity does it use and what is it allowed to do?
AI is no longer limited to answering questions or generating content. Organisations are increasingly experimenting with AI agents that can access applications, interact with APIs, retrieve information, make decisions and execute tasks with limited human intervention. That creates a new cybersecurity question: if an AI agent can access your systems, what identity does it use and what is it allowed to do?
This question is becoming increasingly important as enterprises move from generative AI experimentation towards agentic AI.
The Rise of the Non-Human User
Traditional Identity and Access Management (IAM) has largely been built around people, applications and established machine identities. AI agents introduce another layer.
An agent may be able to:
Access enterprise applications
Retrieve sensitive information
Interact with APIs
Initiate workflows
Make decisions based on available data
Perform actions on behalf of an employee or another system
The challenge isn't simply giving an AI agent access. The challenge is ensuring that the organisation knows which agent is acting, what it is authorised to access, what it is allowed to do and how its actions can be traced.
NIST's National Cybersecurity Centre of Excellence specifically highlighted this emerging challenge in February 2026, announcing work focused on applying identity standards and best practices to software and AI agents. The initiative identifies identification, authorisation, auditing and non-repudiation as key areas requiring attention.
Why Does Traditional IAM Need to Evolve?
Imagine an AI agent connected to your CRM, cloud environment and internal knowledge systems. It may legitimately need access to some of these resources. But should it have unrestricted access? Should its permissions remain the same when its task changes? What happens if the agent's credentials are compromised? And can your security team distinguish between legitimate automated activity and malicious use of the same identity?
These questions become particularly important because AI agents can operate at a speed and scale that traditional human-centric controls weren't necessarily designed around. Gartner's 2026 cybersecurity research identifies Identity and Access Management adapting to AI agents as a major trend, highlighting challenges around identity registration and governance, credential automation and policy-driven authorisation for machine actors.
The Principle of Least Privilege Becomes Even More Important
One of the fundamental principles of cybersecurity is least privilege: give an identity only the access required to perform its legitimate function. For AI agents, this principle becomes increasingly relevant.
Instead of asking: “What can this agent access?” Security teams should be asking: “What does this agent need to access for this specific task?”
That means organisations should consider:
1. Agent identity — every AI agent should be identifiable rather than treated as an anonymous automation process.
2. Controlled permissions — access should be limited according to the agent's role and purpose.
3. Continuous monitoring — agent activity should be logged and monitored just like other important identities and systems.
4. Clear authorisation — organisations need policies defining which actions an agent can execute independently and which require human approval.
5. Auditability — when an AI agent performs an action, organisations should be able to establish what happened, when it happened and under whose authority it operated.
AI Governance Is Becoming Cybersecurity Governance
The conversation around AI security often focuses on model security, prompt injection, data leakage and hallucinations. Those risks matter. But the identity of the AI agent is another critical part of the security equation. An agent with access to sensitive systems can potentially become an extension of the organisation's digital identity infrastructure.
That means AI adoption cannot be separated from identity, access, governance, monitoring and accountability. As Gartner notes, the increasing use of AI agents introduces new attack surfaces and makes cybersecurity oversight essential.
The Question Businesses Should Be Asking
The question is no longer simply: “Are we securing our AI?” It is becoming: “Are we securing everything our AI is allowed to access?”
As organisations adopt agentic AI, establishing clear identity, access and governance controls will become an increasingly important part of responsible enterprise security.
AI can operate autonomously. Security controls shouldn't.
References: NIST — National Cybersecurity Center of Excellence: New Concept Paper on Identity and Authority of Software Agents, February 5, 2026. Gartner: Gartner Identifies the Top Cybersecurity Trends for 2026, February 5, 2026.