Understanding the MITRE AI maturity model: a framework for assessing organizational AI capabilities

Enterprise AI adoption has outpaced the controls meant to govern it. Employees adopt tools ahead of formal approval, and autonomous agents or customer-facing chatbots can reach business areas before formal review. Many organizations struggle to answer a basic question from their board: how mature is our AI program, and which gaps need attention before AI scales?

The MITRE AI maturity model helps leaders make that question concrete. Without a shared maturity baseline, organizations can scale AI pilots and agentic or customer-facing applications faster than governance and data controls can keep pace with in real time.

Use this guide to understand the MITRE AI maturity model, how it structures an organizational assessment, and how it relates to frameworks like NIST AI RMF and the EU AI Act. It also shows where maturity scoring stops short of runtime controls.

Key takeaways

• The MITRE AI maturity model is best used as an enterprise readiness diagnostic that shows where governance, data, technology, and organizational capabilities need investment before AI scales.
• Its six pillars, 20 dimensions, and five cumulative levels turn qualitative assessment into a repeatable roadmap, supported by a cross-functional review and scoring tool.
• MITRE complements frameworks rather than replacing them: NIST AI RMF guides system-level risk management, ISO/IEC 42001 defines certifiable AI management requirements, and the EU AI Act imposes risk-tiered legal obligations.
• Maturity scoring indicates readiness, but runtime enforcement is still needed when prompts, responses, and agent actions can pose real-time risk.

What is the MITRE AI maturity model?

The MITRE AI maturity model (AI MM) helps organizations assess their readiness to adopt AI and identify areas for improvement. It’s a qualitative measurement and benchmarking tool.

The framework assesses an organization’s current AI capabilities and AI’s potential strategic business impact so leaders can prioritize investments toward the capabilities needed for AI readiness.

MITRE, a not-for-profit technology organization, published the guide in November 2023. The model was developed from a systematic review of commercial maturity models and Capability Maturity Model Integration (CMMI) appraisal processes from Carnegie Mellon University.

It also draws from the National Institute of Standards and Technology’s (NIST) AI standards. That CMMI lineage matters because it frames AI adoption as a staged, cumulative progression with each level building on the last.

Structurally, the MITRE AI maturity model organizes around six pillars, 20 dimensions, and five maturity levels. The pillars and levels bring structure, while the dimensions give teams the criteria they use to judge progress.

FOR DEVELOPERS

Let Your Dev Teams Use AI Without Putting Your IP at Risk.

WitnessAI protects source code and credentials in real time, routes sensitive queries to secure internal models, and gives security teams full visibility — without slowing developers down.

Learn More About WitnessAI For Developers →

The six pillars and five maturity levels that structure the assessment

The model measures maturity across six organizational pillars, each broken into dimensions that describe advancing mastery. The pillars define what enterprise AI readiness looks like, while the five levels show how organizations progress through it. The sections below break down each in turn.

The six pillars and what each measures

Each pillar measures a different part of enterprise AI readiness. The assessment tool presents one multiple-choice question per dimension, and the selected answer determines that dimension’s maturity level.

• Pillar 1, Ethical, Equitable, and Responsible Use: This pillar establishes expectations, requirements, and governance to mitigate unintended consequences. It covers responsible and contestable AI requirements, including transparency and human-centric fairness.
• Pillar 2, Strategy and Resources: This pillar measures whether the AI strategic plan has the partnerships and governance needed to support it. It shows whether the organization has connected AI adoption to resources and business priorities.
• Pillar 3, Organization: This pillar evaluates whether the organization’s culture and operating model can build the workforce skills AI requires. It shows whether people and operating models are ready to support AI beyond isolated pilots.
• Pillar 4, Technology foundation: This pillar spans AI innovation, test and evaluation, platform capabilities, and architecture. Its Security and Privacy dimension links directly to AI-native security concerns.
• Pillar 5, Data: This pillar covers architecture, security and privacy, data governance, and accessibility. It assesses whether AI systems have the data foundation needed for reliable, governed use.
• Pillar 6, Performance and Application: This pillar measures usage, monitoring, resilience, and user trust. It focuses on whether AI is delivering value in a way the organization can monitor and improve.

The six-pillar structure helps teams discuss AI maturity across the functions responsible for policy, data, security, and business outcomes without reducing readiness to a single technical score.

How organizations progress through the five maturity levels

Advancement is cumulative: an organization reaches a level only after meeting the benchmarks of the level below. The five levels describe a hierarchical progression in AI adoption.

• Level 1, Initial: This level describes nascent efforts with no AI sponsor, governance, or strategy. AI activity exists, but the organization has not established the structure to manage it consistently.
• Level 2, Adopted: This level marks pilots with rudimentary, project-level governance. AI deployments remain decentralized, which limits repeatability and enterprise oversight.
• Level 3, Defined: At this level, approved enterprise-wide approaches, resources, and processes are documented, with governance, culture, and leadership in place.
• Level 4, Managed: This level means initiatives follow policy and technical standards. Metrics begin to drive decisions instead of informal judgment.
• Level 5, final maturity level: This level describes continuous improvement, high adoption, and leadership updating policy based on quantitative analysis. The organization uses evidence to refine how AI is governed and scaled.

Organizations can set different target maturity levels by mission, resources, and business practices. A financial services firm may target Level 4 in Security and Privacy while accepting Level 3 in AI Innovation.

What an AI maturity assessment surfaces about ungoverned adoption

A maturity assessment shows where governance, data, and monitoring need additional support before AI adoption can scale further. A low score on the Technology or Data pillars is a proxy for concrete exposure already accumulating across the enterprise.

Shadow AI adoption is the clearest example. Employees adopt unapproved AI tools when there is no sanctioned, safe path for using AI, and organizations at lower maturity levels typically lack both a governance policy and the technical monitoring infrastructure to detect it.

Runtime behavior adds another layer. Even sanctioned tools can be manipulated at the point of use, whether through prompt injection attacks, unintended data disclosure, or unauthorized commands sent to connected systems. A maturity model measures organizational readiness, but enforcement still has to happen at the moment an AI system acts. That is the main gap for enterprises deploying autonomous agents.

The emergence of agents with access to tools and multi-step action chains exposes the limits of organizational-level governance. An arXiv analysis of governance found that standards such as ISO/IEC 42001 and NIST AI RMF provide governance intent and accountability, while executable runtime policy must still be implemented in the workflow.

For agentic systems, the question is whether the next specific action is authorized under current policy, identity, approval state, data boundaries, and budget constraints. In December 2025, OWASP formalized this with the Top 10 for Agentic Applications, covering agent-specific risks like identity abuse and tool misuse.

PROTECT

Runtime AI Threats Need Runtime Defense.

WitnessAI’s enterprise AI firewall delivers bidirectional runtime defense, blocking prompt injections, jailbreaks, and data exfiltration before they reach your models or your customers.

Explore Protect →

How the MITRE model relates to NIST AI RMF, ISO/IEC 42001, and the EU AI Act

Use the MITRE model as a diagnostic and roadmap tool before choosing a risk management framework or legal compliance path. It answers “how mature are we?” where other frameworks answer “how do we manage risk?” or “what does the law require?”

The major frameworks work at different altitudes, so mature programs often layer them.

• MITRE AI maturity model: MITRE assesses enterprise-wide adoption readiness across pillars, dimensions, and levels. It helps leaders understand where the organization stands before committing to a risk framework, certification path, or regulatory program.
• NIST AI RMF: The NIST AI RMF organizes activity around four functions, Govern, Map, Measure, and Manage, applied at the AI system level. MITRE built its model with deliberate alignment to NIST standards, but the two operate at different altitudes.
• ISO/IEC 42001: ISO/IEC 42001 specifies certifiable requirements for an AI management system. It also addresses organizational governance, but it defines auditable requirements, whereas the MITRE model offers a qualitative progression of maturity.
• EU AI Act: The EU AI Act imposes mandatory, product-level legal obligations tiered by risk. Serious violations can carry penalties of up to €35 million or 7% of global annual turnover.

Many programs use NIST AI RMF as the operating model inside an ISO/IEC 42001 management system. Organizations can then apply that combination to meet EU AI Act obligations. Use the MITRE model to diagnose where the organization stands before choosing among those paths.

Its CMMI lineage helps teams prepare for compliance work. An organization below Level 3 (Defined) is unlikely to have the documented processes and inventories that certification and high-risk obligations require.

A high maturity posture can still leave risk-management gaps. Organizations can score high on maturity while still lacking a rigorous AI risk management framework comparable to financial or cyber risk regimes. MITRE’s public summaries don’t provide an explicit mapping to EU AI Act risk categories or NIST AI RMF functions. A strong maturity posture can support compliance work, but it doesn’t close every risk-management gap.

FOR COMPLIANCE

What Does AI Compliance Look Like?

WitnessAI automatically logs every AI interaction, masks sensitive data in real time, and enforces regulatory policies across every region and business line. Audit-ready from day one.

See WitnessAI For Compliance →

How to conduct a MITRE AI maturity model assessment

Once you understand the exposure that shows up at low maturity and how the model relates to adjacent frameworks, the next step is running the assessment itself. It follows a five-step process that turns qualitative judgment into a scored roadmap, and it’s designed to be repeated over time so the output is a trajectory rather than a snapshot.

1. Prepare and define scope. Review the model and decide which department, project, or business unit is in view, which pillars matter most, and what inputs should be gathered. Scope also involves setting target maturity levels for each dimension. A regulated business may target Level 4 in Security and Privacy while accepting Level 3 in AI Innovation, and naming that up front keeps the assessment focused.
2. Assemble a cross-functional team. Identify reviewers who can speak credibly to each pillar. That usually means data science and MLOps for the Technology and Performance pillars, legal and compliance for Ethical Use and Strategy, security and privacy for Data, and an HR or workforce lead for Organization. A single-function team almost always over-scores its own area and under-scores the rest.
3. Answer the 20 multiple-choice questions. Work through the tool one dimension at a time, with the relevant reviewer leading each answer and the rest of the group challenging it. Completing the full set generates a score and a visualization of results. Where reviewers disagree, capture the evidence behind each position rather than averaging to a middle answer, because the disagreement itself often points to the real gap.
4. Develop a plan from the visualization. Identify the dimensions furthest from their target level and prioritize investment accordingly. Group the gaps into themes such as governance policy, technical monitoring, or workforce enablement so the roadmap maps to how work actually gets funded, not just to the pillar structure. Assign an owner and a target date to each priority.
5. Repeat on a set cadence. Run the assessment periodically, typically every six to twelve months, to measure progress and adjust the roadmap as conditions change. Agentic adoption, new regulations, and shifts in the vendor stack can all move the target level for a dimension between cycles.

Organizations can request the assessment tool by contacting AIMM@mitre.org, and an interactive version is available at aimaturitymodel.mitre.org. The full text of the 20 questions and the exact scoring algorithm live inside those resources rather than in the public summaries.

Run this way, the assessment gives teams a repeatable way to show a board or regulator where the AI program stands, what to fix next, and who owns each priority.

CONTROL

Can You Prove How Your Organization Governs AI?

WitnessAI generates granular audit trails, enforces policies across every role and region, and redacts sensitive data before it ever leaves your network. Compliance-ready from day one.

See How Control Works →

Advancing maturity with visibility, intent, and runtime controls

Advancing from one maturity level to the next benefits from closing the specific capability gaps the assessment surfaces. Several of those gaps map directly to controls that operate at the point of AI interaction.

The biggest jump usually occurs between the pilot stage and the point at which AI becomes a documented, enforceable way of working across the enterprise. If you’re already sitting on a portfolio of decentralized pilots, you’ve felt this transition firsthand: the controls that were fine for a handful of experiments start to strain as usage spreads across teams, tools, and agents.

The capabilities at that transition include:

• Visibility: Shadow AI poses a distinct detection challenge because it interacts with sensitive information via inference rather than through traditional file transfers. WitnessAI’s Observe moduleprovides network-level discovery of AI applications, AI agents, and AI infrastructure—including MCP server activity where applicable—without requiring endpoint clients or browser extensions.
• Intent-based enforcement: Keyword- and regex-based approaches struggle to identify risk in conversational AI environments, where users rarely use the words a pattern matcher expects. WitnessAI’s Control module applies intent-based policies powered by custom ML models that analyze conversational context and purpose
• Runtime defense: WitnessAI’s Protect module provides bidirectional protection that inspects prompts before they reach a model and filters responses before they reach a user in protected workflows. It delivers 99.7% true positive guardrail efficacy, validated in production customer environments, with consistent protection across more than 100 LLM types.

Control supports a four-action policy enforcement model of allow, warn, block, and route. At policy boundaries, users can receive a warning rather than a hard block, and sensitive queries can be routed to an approved internal model.

Used together, these capabilities give a cross-functional steering committee a shared framework to advance maturity with evidence rather than guesswork. That committee can include legal and compliance leaders along with HR, security, and the AI function.

Building the confidence layer for enterprise AI

The MITRE AI maturity model gives enterprise leaders a defensible way to measure where their AI program stands and to chart a credible path forward. Its value is diagnostic: it surfaces the governance, data, and technology gaps that accumulate as adoption outpaces control.

A maturity score measures readiness; enforcement still needs to occur within the AI workflow. Shadow AI and prompt injection risks are addressed only when enforcement operates at the moment AI acts, and ungoverned agents need the same runtime approach.

WitnessAI’s unified platform combines network-level visibility, intent-based control, and runtime protection to help organizations address the gaps surfaced by the maturity assessment. Its intent-based policies help security and AI teams adopt AI with clearer governance, stronger controls, and actionable evidence.For leaders who need to prove AI governance to a board or a regulator, the next step is seeing that enforcement in your own environment.

The same applies to leaders governing the autonomous agent workforce as agent use expands. Schedule a demo to see how WitnessAI supports secure AI adoption across your organization.