TL;DR: MITRE’s AI maturity model gives organisations a structured way to score readiness across governance, data, technology, and operating model, but WitnessAI’s analysis makes clear that maturity alone cannot govern prompts, responses, or agent actions in real time. That gap matters because AI adoption is already outpacing control design, making runtime enforcement the decisive layer as agentic and customer-facing systems scale.
At a glance
What this is: This is an analysis of MITRE’s AI maturity model showing that enterprise AI readiness can be measured, but real-time control of prompts and agent actions still has to happen elsewhere.
Why it matters: IAM, AI security, and governance teams need to treat maturity scoring as a planning tool, not a runtime control, because autonomous and customer-facing AI can create risk after the assessment is complete.
Context
Enterprise AI programmes often outgrow governance before they outgrow ambition. A maturity model can show where policy, data, operating model, and technology are lagging, but it does not by itself control what an AI system does once a user prompt or agent action is in flight.
That distinction matters most where AI reaches sensitive business processes, customer interactions, or delegated tool use. The article’s core point is that organisational readiness and execution-layer control are different problems, and the second becomes the limiting factor as AI scales.
Key questions
Q: What breaks when AI maturity assessments are used as a substitute for runtime control?
A: The assessment becomes a snapshot of readiness rather than a live control, so unauthorised prompts, tool calls, and agent actions can still occur outside policy. Organisations then overestimate their ability to govern AI because the score reflects structure and process, not enforcement at the moment of execution.
Q: When should organisations prioritise runtime controls over policy documentation for AI agents?
A: They should prioritise runtime controls whenever an agent can access data, invoke tools, or trigger external actions without a person approving each step. Policy documentation still matters, but it does not prevent agent behaviour. If the system can act before review, enforcement has to happen at the moment of execution.
Q: What are the signs that an AI programme is mature on paper but weak in practice?
A: Common signs include multiple approved-looking policies, limited live monitoring, unclear ownership for agent access, and no reliable way to explain what an AI system did during a session. Another warning sign is heavy reliance on manual review after the fact. That usually means the organisation has a governance story, but not an operational control model.
Q: How should security teams govern AI agents that can access enterprise systems?
A: Security teams should govern AI agents as non-human identities with explicit ownership, scoped privileges, and continuous monitoring. The control set should include inventory, task-bound credentials, audit trails, and revocation paths. If an agent can call tools or touch production systems, it belongs in the same governance model as service accounts and other machine identities.
Technical breakdown
How MITRE’s AI maturity model scores enterprise readiness
The MITRE AI maturity model is a qualitative assessment framework that scores an organisation across six pillars, 20 dimensions, and five cumulative levels. It is designed to show whether governance, resources, operating model, technology, data, and performance practices are ready for broader AI adoption. Because the model is cumulative, a team cannot claim a higher level without satisfying the lower one first. That makes it useful as a roadmap and benchmarking tool, but it remains a management assessment rather than a control system. The model helps answer whether the enterprise is prepared to scale AI, not whether each AI action is authorised in context.
Practical implication: Use maturity scoring to prioritise programme investment, but do not confuse it with enforcement at the point of AI use.
Why runtime controls are different from maturity scoring
Runtime controls operate when the model is actually being used. That is where prompt injection, unintended disclosure, unsafe tool calls, and agent delegation decisions can produce harm even in a mature organisation. A maturity model can confirm that policy exists, but it cannot stop a model from revealing data or an agent from taking a disallowed action during a live session. In practice, runtime controls need to inspect prompts, responses, context, identity, and allowed actions as the transaction occurs. That is why governance artefacts and runtime policy are complementary layers, not substitutes for one another.
Practical implication: Place policy enforcement inside the AI workflow, not only in governance documents or periodic assessments.
Why agentic AI exposes the limits of organisational assessments
Agentic systems add tool use, multi-step action chains, and delegated decision-making, which changes the risk profile materially. A board-level maturity score can show whether the organisation has a strategy for AI, but it cannot verify whether a specific next action is authorised under current policy, identity, approval state, or data boundary. That is the gap the article highlights: an enterprise can look mature on paper while still lacking executable controls for autonomous behaviour. For identity teams, the issue is especially sharp because authorisation has to move from a static review model to an event-driven decision model.
Practical implication: Treat agentic AI as an execution-layer authorisation problem, not only a governance maturity problem.
Threat narrative
Attacker objective: The objective is to influence AI-driven decisions or extract value from AI sessions before governance and runtime controls can intervene.
- Entry occurs when employees adopt AI tools ahead of formal approval or when customer-facing systems are exposed before governance catches up.
- Credential or control abuse follows when prompts, tool access, or delegated agent permissions are allowed to operate without runtime policy checks.
- Escalation happens when the model or agent can reach business data, connected systems, or sensitive workflows beyond the original intended scope.
- Impact is governed by whether the organisation can prevent disclosure, unsafe commands, or uncontrolled autonomous actions in real time.
Breaches seen in the wild
- Meta AI Instagram Account Takeover: 20,225 Instagram accounts hijacked via compromised Meta AI support chatbot with overprivileged access.
- OmniGPT breach claim 2025: A hacker claims to have leaked 34 million OmniGPT AI chat messages holding users' API keys and credentials; OmniGPT has not confirmed it.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
MITRE-style maturity scoring is necessary, but it is not sufficient. It gives leadership a common language for readiness, which is useful in boards, audit conversations, and programme planning. But readiness language does not stop a live model from disclosing data or acting on a bad prompt, so governance and runtime control must be treated as separate layers. The practitioner conclusion is simple: score maturity to plan, but govern execution to control risk.
Runtime enforcement is the real control boundary for agentic AI. Once an AI system can take actions, the decisive question becomes whether the next action is authorised in context, not whether the organisation has a policy statement. That shifts control from periodic review to event-time policy evaluation across identity, data, and tool access. Teams should therefore treat agentic workflows as identity-bound execution paths, not just AI features.
AI maturity debt emerges when organisations confuse documentation with operability. Many programmes can describe their AI governance, yet still lack the monitoring, approval, and access logic required to contain live AI behaviour. The gap widens as shadow AI and sanctioned agents converge in the same enterprise environment. Practitioners should measure whether governance can be enforced where the system acts, not just where the programme is described.
Identity governance now extends into AI decision-making paths. If an AI agent can query systems, trigger workflows, or retrieve sensitive content, its privileges need the same scrutiny as any other high-risk non-human identity. Maturity frameworks help identify that requirement, but they do not implement it. The field should therefore align AI governance with identity governance, especially where delegated access and tool use meet real-time business operations.
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.
- Companies using AI governance put over 12 times more AI projects into production, according to a Databricks report.
- Read next: AI Agent Authorisation Guide
What this signals
Agentic AI governance debt: this is the gap between knowing an AI programme is immature and being able to stop risky behaviour in production. As AI adoption spreads, the control question moves from assessment to execution, because boards do not absorb risk from scorecards, they absorb it from incidents and exposed workflows.
The strongest programmes will separate enterprise readiness, system-level risk management, and runtime enforcement instead of treating them as one layer. That is where NIST AI RMF, ISO/IEC 42001, and identity-aware policy enforcement complement each other in practice, especially once agents can act across tools and data boundaries.
For practitioners
- Map AI maturity scores to control gaps Use the maturity assessment to identify where governance, data, technology, and operating-model weaknesses are concentrated, then convert those gaps into funded remediation work.
- Separate readiness from runtime enforcement Keep maturity scoring as a planning and benchmarking exercise, while implementing live policy checks for prompts, responses, and agent actions inside the workflow.
- Inventory sanctioned and shadow AI use Establish visibility into approved tools, unsanctioned tools, and agent activity so the organisation can see where AI is already operating without formal control.
- Review delegated AI access as non-human identity governance Treat agent permissions, connected tools, and service credentials as identity controls that need ownership, scope limits, and revocation paths.
- Pair maturity reviews with incident-ready observability Ensure AI logging, audit trails, and response playbooks can explain what an AI system saw, did, and changed when a live event occurs.
Key takeaways
- MITRE-style maturity scoring helps leaders locate AI governance and operating-model gaps, but it does not control live prompts or agent actions.
- The central risk is not the score itself, but the false comfort created when organisations treat readiness as enforcement.
- Security teams need runtime policy, observability, and identity-bound authorisation if AI systems can touch data or trigger actions.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10 addresses the attack surface, NIST AI RMF and NIST CSF 2.0 set the technical controls, and ISO/IEC 42001:2023 and EU AI Act define the regulatory obligations.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST AI RMF | GOVERN — AI Governance and Accountability | The article contrasts enterprise readiness scoring with AI risk management and accountability. |
| Recommendation — Use GOVERN to define AI governance ownership before scaling maturity targets or deployment scope. | ||
| ISO/IEC 42001:2023 | Clause 4 — Context of the organization | The article positions maturity as a management-system question, not just a technical one. |
| Recommendation — Align AI maturity assessment with organisational context, scope, and governance requirements. | ||
| EU AI Act | Art. 9 — Risk Management System | The article links maturity scoring to legal obligations that need a risk system, not only a score. |
| Recommendation — Map AI maturity findings to a formal risk management system before high-risk deployment. | ||
| OWASP Agentic AI Top 10 | ASI03 — Identity & Privilege Abuse | The article explicitly discusses agentic systems, tool access, and runtime authorisation risk. |
| Recommendation — Apply ASI03 controls to constrain agent privileges and verify each action at execution time. | ||
| NIST CSF 2.0 | PR.AA-05 — Access Permissions, Entitlements and Authorizations | Runtime AI control depends on governing permissions and entitlements for agents and users. |
| Recommendation — Enforce PR.AA-05 by scoping AI and agent access to the minimum required permissions. | ||
Key terms
- AI maturity model: A structured framework for assessing how far an organisation has progressed in adopting and operationalising AI. In practice, it helps teams compare pilots, production use, and enterprise-wide integration by looking at governance, data quality, capability, and lifecycle discipline rather than adoption hype.
- Runtime control: Controls that enforce policy while an AI system is operating, rather than after the fact. For healthcare chatbots, runtime control includes data masking, output filtering, access scoping, and immutable logging so the organisation can defend the interaction itself.
- 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.
- 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.
Deepen your knowledge
The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and agentic AI identity in the context of operational control. It helps practitioners connect identity discipline to the realities of autonomous systems and delegated access.
Published by the NHIMG editorial team on August 17, 2026.
Updated on October 12, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org