Monday moves for AI governance: five steps in 90 days

At a glance

What this is: This is a practitioner playbook for operationalising AI governance in 90 days, with the key finding that telemetry, access controls, and reusable assets must come before scale.

Why it matters: It matters to IAM, IGA, PAM, and security architects because agentic AI and NHI governance collapse quickly when intake, logging, and access routing are left to ad hoc team-by-team decisions.

👉 Read C1.ai's 90-day AI governance playbook for identity and access teams

Context

AI governance breaks down first at the intake and visibility layer, not at the policy layer. When projects, agents, and tools enter the enterprise without a single path for review and logging, identity teams lose the ability to answer basic questions about who or what accessed what, when, and through which route.

For NHI, agentic AI, and broader IAM programmes, the practical problem is the same: the governed path has to be simpler and faster than the shadow path. If telemetry, access provisioning, and wrap controls are added too late, the organisation ends up certifying behaviour it never observed and trying to govern systems it never inventoried.

C1.ai frames this as a 90-day operating model rather than a long-range transformation programme. That is typical of the current market reality, where teams are already dealing with AI use, agent access, and unmanaged tool sprawl before a formal identity governance model exists.

Key questions

Q: How should organizations approach the governance of AI agents?

A: Organizations should adopt a governance framework that incorporates continuous visibility, adaptive IAM practices, and stringent policy-based controls. This ensures that all agent actions are tracked, authorized appropriately, and assessed for compliance.

Q: Why do audit logs not solve AI governance by themselves?

A: Audit logs show activity after it happens, but they do not prevent excessive access or unclear delegation. If the AI actor already has broad credentials, logging only improves evidence quality. Teams need entitlement control, session boundaries, and owner accountability alongside logging.

Q: What are the signs that shadow AI is still winning?

A: The clearest sign is when the unmanaged route is faster than the governed one and teams keep using it. Other signals include incomplete audit coverage, direct access to legacy applications, and reusable assets being rebuilt in separate teams instead of published once and consumed broadly.

Q: How do security teams know if AI governance is working?

A: Look for evidence that access decisions are reviewable, permissions are revocable, and exceptions are not becoming permanent. If the team cannot explain who owns an AI workflow, what it can reach, and when its access was last reviewed, governance is incomplete. Control maturity shows up in traceability, not adoption volume.

Technical breakdown

Single-platform AI intake and identity routing

A single-platform AI funnel is a front-door control for projects, agents, and tools. In practice, it creates one intake path, a small set of required attributes, and a visible audit trail before anything is allowed to proliferate. For identity teams, the architectural point is not perfection at intake. It is establishing a stable record of the actor, the owning team, the intended use case, and the policy path before downstream access decisions fragment across departments.

Practical implication: route every new AI initiative through one governed intake path before it receives tooling, credentials, or production access.

Central audit logging before policy enforcement

Telemetry first, rules second is a control-ordering principle. If the organisation cannot answer what agents did, which tools they called, and what data they touched, then policy becomes aspirational rather than enforceable. Central audit logging gives IAM, SIEM, and governance teams the evidence base for review, exception handling, and incident reconstruction. Without that baseline, organisations tend to write policies that cannot be measured against actual runtime behaviour.

Practical implication: establish central audit logging for in-production agents before you rely on policy controls or review cycles.

Identity-aware proxy wrapping for high-blast-radius apps

An identity-aware proxy sits in front of an existing application and governs access without requiring a rewrite of the system underneath. That makes it useful for legacy apps that cannot be rebuilt quickly but still need controlled access, deprecation of old service-account paths, and clearer attribution. The important architectural effect is blast-radius reduction. It creates a managed access layer that can be observed, logged, and eventually replaced, instead of letting direct legacy credentials persist indefinitely.

Practical implication: wrap the highest-blast-radius legacy applications first and remove old service-account paths as the managed layer takes over.

Threat narrative

Attacker objective: The objective is to operate AI workloads outside governed identity and access controls so visibility, accountability, and enforcement never fully converge.

1. Entry begins when AI projects, agents, or tools bypass a single governed intake path and enter the environment through unmanaged channels.
2. Escalation occurs when unlogged or partially logged agent activity prevents security teams from reconstructing what tools were used, what data was accessed, or whether access was appropriate.
3. Impact is governance collapse, where shadow AI persists because the unsafe path remains faster than the governed one and the organisation cannot reliably measure or constrain it.

Breaches seen in the wild

• Deloitte 2025 Breach — Deloitte 2025 breach exposes GitHub credentials and proprietary source code via access control failure.
• Twitter Source Code Breach — Twitter source code leaked to GitHub by insider including authentication systems and configuration credentials.

Read our 52 NHI Breaches Analysis report for a comprehensive view of breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

Single-path intake is a governance control, not just an operating convenience. When AI projects can enter the enterprise through multiple informal channels, ownership fragments before the first access decision is made. That creates a shadow governance layer that IAM cannot certify, because the system of record was never established. The implication is that identity programmes must treat intake as a control point, not a coordination task.

Telemetry-first governance is the right order because policy without evidence is theatre. The post correctly prioritises audit logging before policy, and that sequence matters for agentic systems as much as for NHI estates. If teams cannot reconstruct agent actions in near real time, access review and exception handling lose their evidentiary basis. Practitioners should treat observable runtime behaviour as a prerequisite for governance maturity.

Identity-aware wrapping is a pragmatic way to constrain blast radius while legacy access patterns are retired. The strongest part of this model is that it does not pretend the underlying application can be rebuilt immediately. Instead, it constrains direct paths, removes stale service-account usage, and forces access through a governed layer. That is a sensible transition pattern for NHI-heavy environments that need control now, not after a multi-year migration.

Fastest-path metrics expose the real competition between governed and shadow access. If the managed route is slower than the unmanaged route, users and teams will choose the path of least resistance. That is how shadow AI returns even when policy exists on paper. The named concept here is the fastest-path delta: the gap between safe access and unsafe access that determines whether governance is actually adoptable.

This operating model reinforces a broader identity lesson: control design fails when it ignores human incentives and runtime speed. IAM, PAM, and NHI governance all depend on the governed path being visible, low-friction, and measurable. If the organisation cannot make the secure path the easiest path, the control stack becomes advisory rather than enforceable. Practitioners should read this as a call to design for adoption, not just authority.

From our research:

• 91.6% of secrets remain valid five days after the targeted organisation is notified, showing a critical gap in remediation procedures, according to Ultimate Guide to NHIs.
• From our research: Only 5.7% of organisations have full visibility into their service accounts, according to Ultimate Guide to NHIs.
• That visibility gap is why lifecycle control and offboarding discipline belong in the same programme as runtime governance, as explored in 52 NHI Breaches Analysis.

What this signals

Fast-path governance is now the difference between adoption and drift. If the secure route is slower than the unmanaged route, users will bypass it and the programme will quietly lose authority. The operational test is whether teams can make the governed path both visible and friction-light enough to win against shadow behaviour. In NHI-heavy estates, that principle applies just as much to service accounts and tool access as it does to human approvals.

The post also points toward a broader control pattern for agentic environments: build observable access paths first, then extend policy depth only after you can prove the runtime is being captured. That sequencing aligns with the wider identity lesson in the Ultimate Guide to NHIs, where visibility and offboarding are prerequisites, not afterthoughts.

The named concept here is the fastest-path delta, which is the measurable gap between governed and shadow access. When that gap stays wide, governance loses adoption even if the policy design is technically sound. Practitioners should expect this metric to become a standard board-level signal for AI and NHI programme health.

For practitioners

• Create one AI intake front door Require every AI project, build or buy, to enter through a single submission path with ownership, intended use, and review metadata. Make it the default route before any team can request tools, credentials, or production access.
• Stand up central audit logging first Capture agent activity into one queryable log before writing enforcement rules so you can answer what agents did this quarter in under 60 seconds. Use that evidence base to drive access review, exception handling, and incident reconstruction.
• Wrap the highest-blast-radius applications Place identity-aware proxies in front of the three applications where direct misuse would create the most damage, then date and deprecate old service-account paths as traffic shifts to the governed layer.
• Publish reusable AI governance assets Create one approved agent, one MCP tool wrapper for a high-demand system, and one policy template that other teams can reuse. Track whether at least one external team consumes each asset so reuse becomes measurable.
• Measure the fastest path weekly Track time-to-credential, time-to-first-agent, and time-to-tool-access on a dashboard, then compare the managed and unmanaged routes. Treat any 25 percent gap or worse as a governance failure that needs immediate correction.

Key takeaways

• The post argues that AI governance should begin with intake, logging, and access routing, not with a long policy roadmap.
• The operational risk is not just lack of control, but the repeated failure to make governed access faster than shadow access.
• For IAM and NHI teams, the key shift is treating visibility and adoption metrics as first-class governance controls.

Key terms

• Fastest-Path Delta: The gap between the quickest governed access route and the quickest unmanaged route. In identity programmes, it is a practical measure of whether people and systems will choose the controlled path or route around it when speed matters.
• Identity-aware Proxy: An identity-aware proxy combines routing with authentication and authorization logic. It checks tokens or certificates, applies policy at the edge, and forwards verified identity context to the backend so applications do not have to re-implement security decisions inconsistently.
• Single-Platform AI Funnel: A single intake path through which AI projects, agents, and tools must pass before they receive credentials, tooling, or production access. It creates a consistent review point and a record of ownership, purpose, and approval.
• Shadow AI: AI agents, copilots, or connected tools operating without full visibility or governance from security teams. Shadow AI becomes an identity problem when those systems authenticate with unmanaged tokens, service accounts, or OAuth apps that can reach production resources.

What's in the full article

C1.ai's full post covers the operational detail this analysis intentionally leaves for the source:

• The exact five-move 90-day sequence and the weekly targets attached to each move.
• The role assignments for CIO staff, CAIO, COO, platform engineering, and business-unit owners.
• The dashboard metrics used to measure fastest-path delta across safe and unsafe routes.
• The year-one progression model from initial funnel to reusable assets and quarterly review rhythms.

👉 C1.ai's full post breaks down the five moves, operating targets, and accountability model in more detail.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity security 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.