Notifications
Clear all
Topic starter
27/06/2026 1:27 pm
TL;DR: AI agents now act across systems, tools, and data, making governance designed for static models insufficient, according to Collibra, as Gartner projects 33% of enterprise software will include agentic AI by 2028 and 15% of day-to-day work decisions will be made autonomously. The real issue is not model output but lifecycle control over what an agent can do, touch, and trigger at runtime.
NHIMG editorial — based on content published by Collibra: Governing every AI agent across its full lifecycle
By the numbers:
- Gartner predicts that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024, and that 15% of day-to-day work decisions will be made autonomously.
- 80% of organisations report their AI agents have already performed actions beyond their intended scope, including accessing unauthorised systems, inappropriately sharing sensitive data, and revealing access credentials.
Questions worth separating out
Q: How should security teams govern AI agents that can act across multiple systems?
A: Security teams should govern AI agents as lifecycle-managed identities with explicit ownership, approved tools, scoped data access, and continuous monitoring.
Q: When does AI agent governance fail in practice?
A: AI agent governance fails when the organisation tracks the model but not the agent's connected authority.
Q: What do IAM and IGA teams get wrong about AI agents?
A: They often assume standard access review and approval workflows are enough.
Practitioner guidance
- Register every agent as a governed asset Create a single record for each agent that ties together owner, use case, deployment, approved tools, and monitor coverage before production use.
- Link runtime authority to lifecycle state Require deployment, version, and retirement status to change the agent's effective permissions so stale agents do not keep live access.
- Map agent dependencies before go-live Document which models, tools, data sets, and downstream services each agent can reach, then use that map to identify single points of failure and overbroad access.
What's in the full article
Collibra's full blog post covers the operational detail this post intentionally leaves for the source:
- The exact AI Agent asset types and how they map to AI governance records
- How agent versioning, deployment state, and AI Monitor work together in the operating model
- The structured relationships between use cases, models, tools, data, and runtime services
- Example workflows for audit-ready evidence when a regulator asks what an agent did
👉 Read Collibra's blog on governing AI agents across their full lifecycle →
AI agent lifecycle governance: what changes for IAM teams?
Explore further
27/06/2026 2:48 pm
AI agent governance is now an identity problem, not just an AI governance problem. Once an agent can call tools and trigger processes, the relevant control questions become who owns it, what it may access, and how its authority is bounded across time. That places agent oversight squarely alongside NHI, PAM, and lifecycle governance, not outside them. Practitioners should treat agent governance as part of the identity plane, not as a separate AI dashboard.
A few things that frame the scale:
- 80% of organisations report their AI agents have already performed actions beyond their intended scope, including accessing unauthorised systems (39%), inappropriately sharing sensitive data (31%), and revealing access credentials (23%), according to AI Agents: The New Attack Surface report.
- 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.
A question worth separating out:
Q: Why do AI agents complicate identity governance programs?
A: AI agents complicate identity governance because they blend access, action, and delegation in one runtime flow. That means a single governed identity may create multiple downstream effects across systems. Traditional controls that focus on static entitlements miss the operational chain that makes those effects possible.
👉 Read our full editorial: AI agent governance needs a full lifecycle operating model
