AI readiness depends on governance, data maturity and zero trust

At a glance

What this is: This is a Forrester-based AI readiness report that says organizations need stronger governance, data platforms, security controls and workforce capability before scaling AI, especially as agentic AI increases attack surface and operational risk.

Why it matters: It matters because IAM, data security, application security and Zero Trust controls now shape whether AI programmes can be governed safely at scale, including the identity and privilege model behind agents and machine access.

👉 Read Arcon's analysis of Forrester's CIO guide to AI readiness

Context

AI readiness is not just about model quality or access to compute. It is about whether an organisation can govern data, control privileges, absorb technical debt, and operationalise security fast enough to support AI at scale.

Forrester’s framing matters to identity teams because agentic systems change the control problem: the question is not only who can use AI, but what identities, permissions and trust boundaries the system can exercise on behalf of the business.

Key questions

Q: How should security teams govern agentic AI as it moves into production?

A: Security teams should govern agentic AI as a class of non-human identity, not as a generic application feature. That means assigning ownership, scoping permissions tightly, logging every tool action, and revoking access on a defined lifecycle. Production rollout should require clear approval points for high-risk actions and continuous monitoring for drift.

Q: Why does AI readiness depend on governance rather than model choice?

A: Because model choice does not fix weak data controls, unclear ownership or unmanaged permissions. Readiness depends on whether the organisation can approve use cases, bound data exposure, manage technical debt and enforce risk controls consistently. If those foundations are weak, even a capable model becomes harder to govern safely.

Q: What are the signs that an AI banking programme is not ready for scale?

A: Warning signs include weak data quality, too little structured training data, heavy dependence on legacy systems, and difficulty cleaning or integrating data from multiple sources. If banks cannot explain model outputs or test them against compliance and risk requirements, the programme is not mature enough for broad deployment. Those gaps usually appear before performance problems do.

Q: Should organisations allow AI experimentation before governance is in place?

A: Only inside bounded, monitored paths. Uncontrolled experimentation tends to create shadow AI, while approved sandboxes let teams test tools without exposing sensitive data or bypassing review. The practical choice is not between innovation and control, but between governed and ungoverned adoption.

Technical breakdown

Why agentic AI changes the trust boundary

Agentic AI is different from a static chatbot because it can select actions, call tools, and operate across workflows with some runtime discretion. That shifts the trust boundary from the user interface to the identity and authorization model behind the agent. If the agent can access data, initiate requests, or trigger downstream systems, then prompt injection and adversarial manipulation become control-plane risks, not just content-safety issues. Security teams need to treat the agent as a governed runtime actor with scoped entitlements, auditable decisions, and constrained tool access. See the AI Risk Assessment Guide. See the AI Risk Assessment Guide.

Practical implication: define the agent’s identity, allowed tools, and escalation path before production use.

Governance and data maturity set the ceiling for safe AI adoption

AI programmes fail when organisations try to scale usage before they can govern inputs, outputs, and business impact. Governance is not a review board in isolation; it is the operating system that decides which use cases are acceptable, which data can be exposed, and which controls are mandatory at each risk tier. Data platform maturity matters because unreliable lineage, poor classification, and weak access control turn model deployment into an uncontrolled data problem. In practice, the readiness gap is often visible long before model quality becomes the limiting factor. See the AI Governance Guide. See the AI Governance Guide.

Practical implication: tie AI approval to data classification, risk tiering, and control owners rather than to model enthusiasm.

Zero Trust must extend to AI tool use and machine access

Zero Trust is relevant here because agentic systems depend on dynamic access to data, APIs, and internal services. The principle is not simply to trust less, but to verify every request, limit blast radius, and assume the agent may be manipulated during execution. That means separating human identity from system identity, enforcing least privilege for machine access, and monitoring tool calls as privileged actions. A readiness programme that ignores these control points will inherit the same weaknesses that broke earlier identity and cloud security programmes, only now at machine speed. See the Cloud Security Maturity Model. See the Cloud Security Maturity Model.

Practical implication: apply Zero Trust rules to agent tool calls, service accounts, and downstream API access.

NHI Mgmt Group analysis

AI readiness is now a governance and identity problem, not a model procurement problem. The report’s core message is that organizations win or lose on AI by how well they manage access, data, operational debt and risk ownership. That is especially true once agentic systems can act through tools and delegated permissions. Practitioners should treat AI readiness as a control architecture question, not a vendor selection exercise.

Agentic AI creates a new class of privileged runtime actor. Once an AI system can call tools, retrieve data or trigger actions, traditional application trust assumptions stop being sufficient. The governance gap is not just prompt safety, but the absence of a defined identity, scope and oversight model for the agent itself. This is where agentic AI identity maturity becomes a practical governance lens for teams trying to scale safely.

Technical debt becomes a security multiplier when AI sits on top of weak foundations. The report is right to connect readiness with modernization because fragmented data platforms and inconsistent controls make AI risk harder to bound. In practice, poorly governed data access, duplicate entitlements and unclear ownership create the conditions for misuse, leakage and untraceable automation. The implication is clear: modernize control planes before you accelerate deployment.

Zero Trust is only meaningful for AI if it reaches the tool layer. Many programmes stop at user authentication, but agentic risk lives in the permissions behind the agent. If the system can reach APIs, databases and internal workflows, then every tool invocation needs the same scrutiny as a privileged request. Practitioners should stop thinking about AI as an isolated interface and start governing it as a distributed trust chain.

AI literacy is the missing control that makes every other control usable. Security teams can define policies, but business teams decide whether those policies are followed in design, procurement and operations. When the workforce cannot recognise when an AI use case is too risky, governance slips into retrospective exception handling. The organisation that can scale AI safely is the one that makes risk understanding part of delivery, not a separate compliance afterthought.

From our research library:

• Only 13% of organisations feel extremely prepared for the reality of agentic AI despite the majority racing toward autonomous adoption, according to the 2026 Infrastructure Identity Survey.
• Read next: Agentic AI Identity Maturity Model

What this signals

AI readiness debt: the hidden backlog is not only technical debt, but the accumulated inability to explain which identities, permissions and data paths an AI system can use. That debt grows every time a team ships a pilot without a control owner or a rollback path, and it becomes most visible when agentic behaviour starts crossing workflow boundaries.

The control problem is shifting from static access review to runtime access governance. That means security teams need to watch for tool calls, delegated actions and data retrieval patterns that indicate the AI system is operating outside its intended boundary, not just whether the model is accurate. A programme that cannot observe those events cannot credibly claim it has governed AI risk.

For practitioners

• Define the AI trust boundary Document which identities, tools, data sources and workflows an AI system may access, and require approval before those permissions expand.
• Classify AI use cases by risk tier Assign each use case an explicit risk level, control owner and acceptable data scope before pilot approval.
• Map agent permissions to Zero Trust controls Treat every tool call, API request and delegated action as a privileged event that needs scoping, logging and verification.
• Modernize data access governance Reduce uncontrolled access paths, clarify data ownership and tighten entitlements where AI will train, retrieve or act on sensitive content.

Key takeaways

• AI readiness is governed by operating maturity, not by the novelty of the underlying model.
• Agentic AI raises the stakes because runtime decisions can cross data and tool boundaries through legitimate access paths.
• The practical fix is to bind AI deployment to identity, data and Zero Trust controls before scaling use cases.

Related NHI Mgmt Group resources: Agentic AI Identity Maturity Model · AI Risk Assessment Guide · AI Governance Guide · Cloud Security Maturity Model

Key terms

• AI readiness: AI readiness is the state where an organisation can deploy AI systems without losing control of identity, access, and auditability. It goes beyond adoption or enthusiasm and asks whether the environment can govern AI tools and agents across the full stack, including data, devices, and lifecycle processes.
• 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 access governance: Runtime access governance is the practice of deciding and enforcing access based on current execution context rather than only on static assignment. For autonomous or semi-autonomous actors, it requires time-bound permissions, audit trails, and revocation logic that can keep up with live action.
• Zero Trust for AI: A control model that applies continuous verification and least privilege to AI systems, but extends enforcement into data access and action scope. It recognises that AI risk is not limited to authentication and must include how systems move, transform, and share information.

What's in the full report

Arcon's full report covers the operational detail this post intentionally leaves for the source:

• Forrester's readiness framework and scenario planning guidance for different AI advancement paths
• The report's detailed discussion of governance, technical debt and data platform maturity as readiness factors
• The security and risk control considerations behind agentic AI adoption, including prompt injection and model poisoning
• The report's framing for aligning AI investment with measurable business outcomes

👉 Arcon's full report covers the readiness framework, control priorities and scenario guidance behind the analysis.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security and secrets management. It helps practitioners connect AI access patterns to the identity controls their programmes depend on.