Governing AI agents ahead of regulation: what enterprises need

At a glance

What this is: This is a Zenity event preview arguing that AI agent governance is moving faster than enterprise compliance models, with agentic AI exposing gaps across DLP, model security, and application security.

Why it matters: It matters because IAM, IGA, PAM, and security teams need to prepare for runtime governance of agents now, before regulation and operational controls diverge further.

Context

AI agent regulation is becoming an identity governance problem, not just a policy discussion. The article focuses on agentic AI deployment, where runtime decision-making, tool use, and delegated access cut across DLP, model security, and application security.

The core gap is that existing control models still assume bounded workflows and stable authorisation paths. When an AI agent can choose actions at runtime, the governance question shifts from who approved a request to what authority the agent can exercise while it is executing.

The event is timed around July 30, 2026, but the underlying issue is not date-specific: enterprises are being forced to plan for compliance before the regulatory picture fully settles.

Key questions

Q: What breaks when AI agents are not governed at runtime?

A: Without runtime governance, an agent can shift behaviour after provisioning and still execute actions that were never reviewed in context. That is where tool chaining, MCP connections, and rapid decision-making become dangerous. Static approval cannot stop a live change in intent, so teams lose control at the point of action.

Q: Why are regulators focusing on agentic AI security now?

A: Regulators are reacting to a control gap, not a hype cycle. Agentic AI can cross data, model, and application boundaries while acting with delegated authority, so policy makers are moving toward stronger accountability, hardening, and cybercrime enforcement.

Q: Who is accountable for AI agent governance under compliance frameworks?

A: Accountability sits with the organisation that allows AI agents to access data, not with the model itself. Boards, CISOs, and control owners need to show who approved access, what data was reachable, what protections were enforced, and how every action was logged. Frameworks such as NIST AI RMF, ISO 42001, and the EU AI Act all expect evidence, not assertions.

Q: How should security teams prepare for AI agent oversight requirements?

A: Security teams should align identity, risk, and application controls around agent behaviour instead of platform ownership alone. That makes it easier to prove who controls execution, who reviews exceptions, and what evidence exists if rules tighten.

Background and context

Why agentic AI breaks bounded-governance assumptions

Agentic AI changes the control problem because the subject of governance is no longer a static application, but a runtime actor that can select tools, sequence actions, and continue execution without a human approving each step. That creates a mismatch with governance models built around fixed entitlements and predictable request flows. DLP, model security, and application security each cover part of the stack, but none of them fully describe the authority of the agent itself. The practical issue is not simply access volume, but where decision authority lives during execution.

Practical implication: Treat the agent as a governed identity layer, not just an application feature.

Why compliance teams are watching federal AI policy signals

The article frames the Trump Administration's AI Executive Order as a policy signal, not a complete compliance answer. The relevant point for practitioners is that federal hardening, frontier model developer frameworks, and stronger enforcement against AI-enabled cybercrime all point toward more explicit expectations around control, accountability, and deployment discipline. That matters because enterprises rarely get to design compliance from scratch once regulation arrives. They usually inherit obligations that are easiest to satisfy when identity boundaries, logging, and delegated authority are already mapped.

Practical implication: Use current federal policy direction to pressure-test your agent governance model before formal obligations land.

Where DLP, model security, and application security stop being enough

Agentic AI cuts across three control planes at once. DLP looks at data movement, model security looks at model behaviour and exposure, and application security looks at the software surface. An autonomous or semi-autonomous agent can sit in the overlaps between them and still remain insufficiently governed if no one owns the delegated authority layer. That is why agent governance is becoming a separate discipline rather than a subtopic of existing tooling. The missing control is often not detection, but explicit lifecycle ownership of what the agent can do at runtime.

Practical implication: Map one owner to the agent's runtime authority across data, model, and application controls.

NHI Mgmt Group analysis

Agentic AI governance is becoming a distinct control plane, not a feature of adjacent security tools. DLP, model security, and application security each address a slice of the risk, but the article shows that none of them fully govern runtime decision-making by agents. Once an AI agent can choose actions and execute them dynamically, the security problem moves to authority management. Practitioners should treat agent governance as its own operating model.

Compliance will increasingly fail at the boundary between policy intent and execution authority. Existing governance approaches were designed for bounded tools and predictable authorisation paths. That assumption breaks when an agent can act within changing context and assemble its own sequence of steps. The implication is that identity governance must move closer to execution time, because retrospective review alone will not explain what the agent was permitted to do.

Federal AI policy is signalling the direction, but enterprise readiness will still depend on local control design. The executive-order framing points to hardening, accountability, and stronger cybercrime enforcement, which will shape expectations even before sector-specific rules settle. The article therefore signals a market shift toward provable control over agent behaviour rather than broad policy statements. Security teams should expect scrutiny of delegated access, logging, and ownership.

Runtime authority is the named concept enterprises need to manage now. The problem is not simply that AI agents exist, but that they exercise authority while acting, which makes their identity posture materially different from conventional software. That distinction is becoming the dividing line between aspirational governance and enforceable control. Teams that cannot explain runtime authority will struggle to defend their programme under future regulation.

Identity governance for agents will converge with compliance, risk, and security policy faster than most programmes expect. The article points to a future in which regulatory language, control ownership, and agent execution design become tightly linked. That means the identity team, the AI risk team, and application security can no longer work as separate silos. Practitioners should align those functions before policy pressure forces a rushed redesign.

From our research library:

• Organisations that describe themselves as confident in their AI deployment actually experience a 72% security incident rate, compared to 33% for those who remain cautious, according to the 2026 Infrastructure Identity Survey.
• 67% of organisations still rely heavily on static credentials despite the risks they pose to agentic AI deployments, according to the 2026 Infrastructure Identity Survey.
• Read next: Agentic AI Identity Guide

What this signals

Runtime authority is the control gap enterprises need to name. Agentic AI does not just consume access, it can actively shape the sequence of access decisions during execution. That means governance must move from static entitlement thinking to authority-in-use thinking, or the programme will always be reviewing the wrong artefact.

The compliance burden will land unevenly across teams unless ownership is explicit. When agent behaviour touches data protection, model risk, and application security at once, organisations need one accountable design for delegated authority. Otherwise each control domain assumes another team is handling the agent's decision layer.

Only 13% of organisations feel extremely prepared for the reality of agentic AI, according to the 2026 Infrastructure Identity Survey. That gap is why readiness conversations now need to cover runtime governance, not just deployment velocity.

For practitioners

• Map agent runtime authority Document what each agent can decide, which tools it can call, and where human approval still exists. Use that inventory to identify where current governance assumes stable access paths that no longer apply.
• Separate agent controls from adjacent security tools Assign ownership for delegated agent behaviour across DLP, model security, and application security instead of leaving the overlap unmanaged. The key is to track who governs the agent's execution authority, not just the surrounding platforms.
• Prepare compliance evidence now Retain logs, approval records, and policy mappings that show how agent decisions are bounded today. That evidence will matter when regulators ask how runtime autonomy was constrained before formal rules were finalised.
• Review frontier-model deployment assumptions Check whether your deployment standards still assume predictable tool use and stable authorisation paths. If they do, update them to account for agents that change action paths during execution.

Key takeaways

• Agentic AI turns governance into a runtime problem because the agent can choose actions, tools, and timing within execution.
• The article frames policy change as an early warning that compliance expectations are moving toward explicit control of delegated authority.
• Enterprises that cannot explain and evidence agent runtime authority will struggle to defend their identity and security programme.

Key terms

• Agentic AI: Autonomous AI systems capable of planning, deciding, and taking actions, including calling APIs, writing code, and orchestrating other agents, with minimal human oversight. Agentic AI introduces new NHI risks as agents must authenticate to external services.
• Runtime authority: Runtime authority is the permission an AI system has while it is actively deciding and acting, not just when it is approved. In governance terms, it is the point where access, tool use, and action scope become operational, which is why build-time review alone cannot prove safety.
• Delegation authority: Delegation authority is the right for one identity to act on behalf of another within a defined scope. For agents, it must be explicit, time-bound, and auditable because the system may initiate actions independently once granted access.
• Compliance Evidence: Compliance evidence is the artefact trail that proves a control operated as intended. In identity programmes, that usually includes approvals, review outcomes, revocation records, and exception handling. Strong evidence is time-bound, attributable, and reusable across audits instead of being rebuilt manually for each framework.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an identity security programme, it is worth exploring.