Pillar 4Governance Produces: Governance & Data Governance Maturity Profile

Governance, Risk, Compliance & Data Governance

Build the oversight, accountability, and data foundations trustworthy AI stands on.

Aligns withNIST · Map / Measure / ManageISO/IEC 42001ISO/IEC 27001HECVATFERPA / HIPAAEU AI Act risk tiers

The idea

This pillar is the structural foundation the others depend on: bodies, decision rights, risk processes, legal compliance, and data governance. It faces two opposite failures: too little oversight, so AI spreads ungoverned, and too much process, so a heavyweight committee approves little and units bypass it. The goal is proportionate governance: more scrutiny for higher-stakes, more sensitive uses, and a lighter path for low-risk ones.

The pillar is defined by concrete structures you can establish and demonstrate: a standing governance body with a charter and decision rights; risk assessment tiered by impact and data sensitivity; a map of obligations under FERPA, ADA, GDPR, and state law; and a data-governance baseline for classification, access, retention, and vendor terms. AI deployed without data governance is a breach waiting to happen.

Why it matters

Governance turns good intentions into reliable behavior, and data governance is non-negotiable: the fastest way to turn an AI initiative into a crisis is to feed sensitive student data into a system nobody vetted. This is where accountability becomes real: a named body, a documented decision, and a defensible answer when a regulator or journalist asks, 'Who approved this, and how?'

Small-unit formKnow which data is sensitive, and keep a simple list of approved tools.

How you can tell

When it's working
  • A standing AI governance body has a charter, faculty and student members, and decision rights.
  • Risk assessment is tiered by impact and data sensitivity, with a fast lane for low-risk uses.
  • Compliance obligations (FERPA, ADA, GDPR, state law) are mapped and checked across the lifecycle.
  • Data is classified, access-controlled, retention-bound, and covered by vetted vendor terms.
When it's missing
  • No one owns AI oversight; approvals depend on whom you ask.
  • Every use gets the same scrutiny (or none), so governance is a bottleneck or a fiction.
  • Compliance gaps surface in incident response, not design.
  • Student data flows into AI tools under unread terms of service.

How to operationalize it

Rate your campus

Each practice is something you can build, assign, and evidence, with a first move, an owner, and what proof looks like. Move one practice up one level at a time.

The five levels: 1 Initial · 2 Emerging · 3 Defined · 4 Managed · 5 Optimizing. Each practice describes levels 1, 3, and 5. Level 2, Emerging: some level-3 behavior exists in pockets, but informally. Level 4, Managed: level-3 behavior is applied consistently, measured, and reviewed on a set cadence.

1

AI governance body & charter

A cross-functional body with defined authority, a charter, and student representation.

Not yet rated
1InitialNo body, or an advisory group with no authority and no charter.
2EmergingGood practice exists in pockets, but it is informal and has not been institutionalized.
3DefinedA chartered body exists with cross-functional membership, defined authority, and student voice.
4ManagedThe practice is applied consistently, measured, and reviewed on a defined cadence.
5OptimizingThe body reports annually to leadership and the board on posture, risks, equity, and sustainability.
Next moves
  1. Charter the body: membership, authority, decision rights, relationship to shared governance.
  2. Include explicit student representation and library/accessibility voices.
  3. Commit to an annual report to cabinet and board covering risk, enablement, equity, and cost.
Who owns it

President / Provost (charter) · Governance chair

Evidence it exists
  • Approved charter
  • Membership roster
  • Annual report to leadership
2

Risk tiering & algorithmic impact assessment

A three-tier risk standard (assistive / operational / consequential) with impact assessments before consequential deployments.

Not yet rated
1InitialEvery AI request is treated the same, or not reviewed at all.
2EmergingGood practice exists in pockets, but it is informal and has not been institutionalized.
3DefinedA risk-tiering standard is adopted; consequential systems require an impact assessment before deployment.
4ManagedThe practice is applied consistently, measured, and reviewed on a defined cadence.
5OptimizingBias audits and impact reviews run on schedule, and findings demonstrably change deployments.
Next moves
  1. Adopt a three-tier standard calibrating requirements to actual risk.
  2. Require an algorithmic impact assessment for any system touching consequential decisions.
  3. Schedule recurring bias audits for deployed consequential systems.
  4. Connect tiering to research review: human-subjects studies that use AI tools or AI-processed data go through the IRB, with AI-specific consent and data-handling questions.
Who owns it

AI governance body · Institutional research · Legal · IRB

Evidence it exists
  • Tiering standard
  • Completed impact assessments
  • Bias audit schedule and findings
  • IRB AI review questions
3

AI system inventory

A maintained register of AI systems: purpose, owner, data, affected populations, risk tier, audit schedule.

Not yet rated
1InitialNobody can list the AI systems touching institutional data.
2EmergingGood practice exists in pockets, but it is informal and has not been institutionalized.
3DefinedAn inventory exists with owner, vendor/model, data categories, risk tier, and oversight point per entry.
4ManagedThe practice is applied consistently, measured, and reviewed on a defined cadence.
5OptimizingInventory entries are created automatically at procurement and deployment; retirement owners are named.
Next moves
  1. Stand up the inventory with the Nine-Domains field set (purpose, owner, data, populations, tier, audit, retirement).
  2. Trigger entries from procurement and deployment approval so the inventory can't drift.
  3. Review the inventory annually alongside the risk-tiering standard.
Who owns it

CIO · CISO

Evidence it exists
  • Inventory with required fields
  • Procurement-trigger workflow
  • Annual review record
4

Data governance for AI

Classification-based rules for what data can meet which tools, plus shadow-AI discovery.

Not yet rated
1InitialInstitutional data flows into unapproved tools with no visibility, shadow AI is unmanaged.
2EmergingGood practice exists in pockets, but it is informal and has not been institutionalized.
3DefinedAn approved-tool list organized by data classification is published and maintained; shadow-AI discovery exists.
4ManagedThe practice is applied consistently, measured, and reviewed on a defined cadence.
5OptimizingDiscovery, remediation, and approval are a managed loop; audit trails cover AI systems processing institutional data.
Next moves
  1. Publish a searchable approved-tool list organized by data classification.
  2. Stand up shadow-AI discovery connected to procurement and governance escalation.
  3. Define audit-trail and logging standards for systems processing institutional data.
Who owns it

CISO · Data governance committee

Evidence it exists
  • Approved-tool list
  • Shadow-AI discovery reports
  • Logging standards
5

Procurement & vendor risk

AI-specific contract provisions, vendor documentation requirements, and a rapid pilot lane.

Not yet rated
1InitialAI tools arrive through P-cards and free trials; contracts are silent on AI.
2EmergingGood practice exists in pockets, but it is informal and has not been institutionalized.
3DefinedContracts require data-use restrictions, provenance, indemnification, and audit rights; a cross-functional review team exists.
4ManagedThe practice is applied consistently, measured, and reviewed on a defined cadence.
5OptimizingA rapid-cycle pilot lane handles low-risk tools on time-limited approvals; legislation tracking feeds contract standards.
Next moves
  1. Adopt AI contract riders: training-data provenance, data-use limits, liability, audit rights.
  2. Convene a cross-functional procurement review (legal, security, privacy, academic, library).
  3. Create a fast lane for low-risk pilots with defined thresholds and expiration dates.
Who owns it

Procurement · Legal · CISO

Evidence it exists
  • Contract rider language
  • Review-team charter
  • Pilot-lane approvals log

Key governance artifacts

Two living records turn governance into oversight. You can't govern, or right-size review for, what you can't see.

AI Use Case Registry

The single source of truth for every AI system in use or proposed.

A living inventory of every AI initiative: production systems, pilots, and proposals. It surfaces shadow AI, enables proportionate review, and answers the question a board or regulator will ask: “How many AI systems touch our students, and who owns them?”

Why it mattersEvery other control (risk assessment, review, incident response, procurement) starts from knowing the system exists.
Each entry captures
Use case ID & nameDescription & purposeApplication domainOwner & unitStatus (proposed / pilot / production / retired)Data sensitivity / classificationVendor or in-houseCompass risk tierDate added · last reviewedLinked risk-registry entries
Get the template

AI Risk Registry

The catalog of identified AI risks, each owned and being managed.

A central record of AI risks (bias, privacy, security, compliance, operational, reputational, academic integrity), each scored for likelihood and impact, owned, and paired with a mitigation. It turns 'we should look into that' into a task with a name on it.

Why it mattersUnrecorded risks get rediscovered during incidents. A registry makes risk visible, ranked, and owned, so mitigation happens before harm.
Each entry captures
Risk IDRelated use caseRisk descriptionCategory (bias / privacy / security / compliance / operational / reputational)Likelihood (1–5)Impact (1–5)Severity (L × I)Mitigation & controlsOwnerStatus (open / mitigating / accepted / closed)Review date
Get the template

In practice

Worked example

Proportionate, not paralyzing

A university's AI committee insists on reviewing every use, including a faculty member drafting discussion prompts from public material. Adoption stalls and units go around it. Under this pillar it adopts tiers: low-risk uses with no sensitive data self-certify against a checklist, medium-risk uses get a short review, and only high-stakes uses touching protected data get the full process. Throughput rises, shadow AI falls, and attention goes where the risk is.

Watch for

  • One-size review that treats a discussion-prompt helper like an admissions algorithm.
  • Governance bodies without decision rights, which advise and are ignored.
  • Treating data governance as IT's problem rather than the foundation of AI risk.

Readiness & maturity questions

Readiness asks: are we prepared? Use these to surface blind spots before you build. Every “no” shows where the work is. Representative prompts for reflection, not a scored test. For the scored version, use the Maturity Assessment.

Governance
  1. Do we have a formal AI governance structure or body?
  2. Are governance roles and responsibilities clearly assigned?
  3. Is there executive-level sponsorship of AI governance?
  4. Are academic, administrative, and student voices represented?
  5. Are governance decisions documented and transparent?
  6. Do committees have access to AI subject-matter expertise?
  7. Are governance practices aligned with institutional mission?
  8. Is there a process to escalate urgent AI-related issues?
  9. Do we regularly review AI use across campus?
  10. Is there coordination between AI governance and IT/data governance?
Risk
  1. Have we defined AI-specific risk categories (e.g., reputational, bias, data)?
  2. Do we use a risk-assessment checklist for AI projects?
  3. Is there a risk scoring or classification framework?
  4. Are high-risk AI uses (e.g., admissions, grading) flagged?
  5. Are stakeholders trained on how to identify AI risks?
  6. Is there a process to evaluate third-party AI vendor risks?
  7. Are risks evaluated at multiple lifecycle stages (design, deployment, monitoring)?
  8. Are there risk thresholds requiring executive or governance approval?
  9. Is incident-response planning integrated with AI operations?
  10. Are risks evaluated across technical, legal, and ethical dimensions?
Compliance
  1. Do we have documented AI compliance requirements (e.g., FERPA, ADA)?
  2. Are AI initiatives reviewed for regulatory alignment?
  3. Are compliance officers involved in AI tool review or procurement?
  4. Are legal and ethical compliance criteria part of project approval?
  5. Are compliance standards built into system design (e.g., privacy by design)?
  6. Are data access and audit controls in place for AI tools?
  7. Are students aware of their rights in AI-mediated environments?
  8. Do policies and terms of use cover AI system behavior and responsibilities?
  9. Is AI-related training tied to compliance expectations?
  10. Are external standards (e.g., EU AI Act, NIST RMF) referenced in planning?

Applicable legislation, policies & standards

Governance doesn't happen in a vacuum, it operationalizes obligations that already bind the institution. Applicability depends on your institution type, jurisdiction, and the data involved; always confirm with your own legal, privacy, and compliance offices.

Legislation & regulation
FERPAPrivacy of student education records, the central constraint on AI touching student data.
ADA & Section 508 / 504Accessibility of AI-enabled tools and digital content.
Title VI & Title IXNon-discrimination, bars AI from producing disparate impact in decisions about people.
GDPREU data protection, where you serve or process EU residents' data.
EU AI ActRisk-tiered obligations for higher-risk AI used in or with the EU.
State privacy laws (CCPA/CPRA & peers)Consumer/data-privacy rights varying by state.
HIPAAWhere AI touches protected health information (clinics, counseling, research).
COPPAWhere systems involve learners under 13 (dual-enrollment, pre-college).
Standards & frameworks
NIST AI RMFGovern · Map · Measure · Manage, the risk-management spine.
ISO/IEC 42001Certifiable AI management-system standard.
ISO/IEC 27001Information-security management.
NIST Cybersecurity & Privacy FrameworksSecurity and privacy control baselines.
HECVATHigher Education Community Vendor Assessment Toolkit, third-party/vendor risk.
OECD AI PrinciplesInternationally recognized trustworthy-AI values.
Institutional policies
Data classification & governance policyDefines sensitivity levels that gate AI data use.
Acceptable use policyWhat AI tools may be used, with what data, by whom.
Records retention & dispositionHow AI inputs, outputs, and logs are retained.
Procurement & vendor risk policyVetting, contracts, and data-processing terms for AI vendors.
IRB / human-subjects policyWhere AI is used in research involving people.

Apply this pillar