AI Governance in K-12: A Practical District Guide

Current status: This guide is educational information, not legal advice. Tenet Gateway is a founding-district program.

K-12 AI governance is the operating system a district uses to decide which AI uses are allowed, under what conditions, for whom, with which data, and with what evidence. It connects board direction to daily choices by educators, students, staff, vendors, and district applications. A policy is part of that system, but policy without ownership, controls, and review is only a statement of intent.

This guide turns that broad idea into a practical district program. It draws on NIST’s voluntary AI Risk Management Framework, federal education privacy resources, and the U.S. Department of Education’s guidance on human-centered educational technology.

Start with a complete definition

A district AI governance program should be able to answer seven questions for every material AI use:

Who or what is acting? A student, educator, staff member, vendor application, or district service.
Under whose authority? A district policy, course decision, job responsibility, contract, or approved application registration.
For what purpose? A specific educational or operational outcome, not a vague statement such as “innovation.”
With which data? Defined systems, records, fields, documents, and data classifications.
Through which deployment? The product, account type, provider, model, region, logging configuration, and contract that actually apply.
Under what constraints? Human review, age limits, prohibited uses, retention rules, accessibility needs, and technical guardrails.
With what evidence? An approval record, policy version, decision event, outcome review, exception, or incident record appropriate to the risk.

If a district cannot answer these questions, it does not yet have a governable use. It has an experiment whose boundaries need to be defined.

Governance covers two planes

Districts encounter AI in two technical forms:

Direct use at the Edge. A person interacts with an AI product. Tenet Edge applies district and classroom guardrails on supported products through a managed Chrome client.
Backend application use. A district or vendor application calls a model through an API. Tenet Gateway is a founding-district program for connecting those operations to district authorization policy.

Both planes use common concepts such as identity, purpose, approved data, model eligibility, policy version, and evidence. Their enforcement points and data paths differ. Read The Two Planes of District AI for the architecture in detail.

The eight parts of an effective district program

1. Leadership and decision rights

Name an accountable executive sponsor and document who can make which decision. Useful roles include:

Decision	Accountable role	Required consultation
District AI principles and risk tolerance	Superintendent or cabinet designee	Board, legal, technology, curriculum, privacy
Instructional approval	Curriculum leader	Educators, accessibility, technology, student services
Data disclosure and privacy model	District privacy or legal lead	Data owner, security, procurement
Technical eligibility and deployment	Technology leader	Security, privacy, application owner
Classroom variation	Designated educator or school leader	Curriculum and district policy owners
Exception and incident response	Named governance group	Relevant operational, legal, and student-support teams

A committee with no decision rights can become a discussion forum. A single technology owner can miss instructional, accessibility, and legal context. The district needs both clear authority and cross-functional review.

2. A living inventory

Inventory more than obvious chatbots. Include:

standalone AI products used by students or staff;
AI features added to software already under contract;
district-built scripts, automations, and applications;
model APIs used by vendors or district teams;
pilots, grants, research projects, and teacher-led experiments;
browser, productivity, communication, assessment, and administrative features that use AI.

Each entry needs an owner, purpose, user group, data boundary, model deployment, approval status, and review date. The district AI application register template provides a starting structure.

3. A risk-tiered approval process

Not every use requires the same review. A staff member generating generic meeting agenda ideas has a different risk profile from an application using education records to recommend student services.

Define tiers using factors the district can observe:

age and role of users;
consequence of the output;
sensitivity and volume of data;
whether the system takes action or only assists a person;
whether a person can review and correct the result;
scale and duration of use;
accessibility and discrimination risks;
ability to stop, export, or reverse the use;
provider, deployment, and contract controls.

A higher tier should lead to stronger evidence, testing, approval, and review. It should not lead to an automatic ban or an automatic approval.

4. Privacy and data governance

Treat “student data” as the beginning of a question, not the end. Document the specific fields and records involved, where they originate, what is sent to the AI service, what the service retains, and who can access outputs and logs.

The FERPA school official exception and COPPA school authorization can be relevant in certain circumstances, but neither is a blanket approval for any tool marketed to schools. Under FERPA, a district must determine whether its particular arrangement satisfies the applicable requirements, including direct control and legitimate educational interest where the school official exception is used. FTC guidance describes limited circumstances in which a school may authorize collection under COPPA for an educational context. Districts should make these determinations with their own counsel and consider state and local rules.

Use K-12 AI data boundaries to document purpose, system, data class, model deployment, retention, and evidence.

5. Instructional quality and human responsibility

An approved tool is not automatically an approved assignment. Instructional governance should define:

the learning goal and the role AI may play;
what students must do themselves;
how sources, uncertainty, and fabricated output are addressed;
when disclosure or citation of AI assistance is expected;
how educators review work fairly;
how students without equivalent access can participate;
how accessibility and language needs are met;
when AI is inappropriate because it would replace the intended learning.

The U.S. Department of Education emphasizes keeping humans in the loop and centering teaching and learning goals. Districts can turn that principle into assignment-level expectations that educators and students can actually use.

6. Technical and procedural controls

Controls should match the plane and the risk. Examples include:

managed access to approved products and account types;
district and classroom guardrails for direct-use surfaces;
application identity and authorization for backend use;
least-privilege access to approved systems and fields;
personal-information redaction on supported direct-use text paths;
provider configuration, retention, and training-use settings;
human approval before consequential action;
documented exception and break-glass procedures;
a tested way to disable an application or revoke a credential.

Tenet Edge currently addresses direct use on supported surfaces. Tenet Gateway is a founding-district program for backend authorization. Neither label removes the need to document the exact supported configuration and the limits of each control.

7. Evidence, incidents, and transparency

Collect the evidence needed to run the program, not an unlimited archive. A district may need to know:

which policy and approval version applied;
whether a use was allowed, denied, constrained, or escalated;
which application or governed identity was involved;
whether an exception was used;
who reviewed an incident and what changed;
when an approval expires.

Define access, location, retention, aggregation, and deletion before collecting an event. Event metadata can still be sensitive. “Aggregated” does not automatically mean anonymous, and a pseudonymous identifier can often be linked back under controlled conditions.

Transparency should include plain-language notices for staff, students, and families that match the actual deployment. Avoid describing future controls as current or implying that governance software replaces educator judgment.

8. Continuous review

AI products, models, terms, and district uses change. Establish both scheduled reviews and material-change triggers.

Review when:

the purpose or eligible users change;
a new data source or data class is added;
the provider changes retention, training use, terms, or subprocessors;
the model or deployment route changes;
an AI feature becomes capable of taking action;
testing reveals a quality, accessibility, privacy, or safety issue;
an incident or complaint exposes a control gap;
the owner leaves or the educational program ends.

A 90-day implementation sequence

Days 1 to 30: make the work visible

Name the executive sponsor and working group.
Publish interim principles for staff and educators.
Build the first inventory from procurement, identity, browser, cloud, and departmental sources.
Identify high-consequence and high-data uses for immediate review.
Establish a simple intake and triage process.

Days 31 to 60: create repeatable decisions

Adopt a common vetting record using the AI tool vetting template.
Define risk tiers and required reviewers.
Publish approved, conditional, pilot, denied, and retired status definitions.
Create data-boundary and deployment-eligibility fields.
Document how exceptions and material changes are handled.

Days 61 to 90: connect policy to practice

Apply technical controls to supported direct-use surfaces.
Register backend applications and their model routes.
Train educators and staff on purpose-specific rules.
Publish plain-language family and student information.
Test evidence, escalation, revocation, and review workflows.
Report program measures to leadership without overstating certainty.

Measures that show whether governance is working

Avoid a single “AI safety score.” Use a small set of operational measures tied to district decisions:

percentage of known AI uses with an accountable owner;
percentage with a documented purpose, data boundary, and review date;
time from intake to a risk-tiered decision;
number of overdue material-change reviews;
percentage of approved uses with tested disable or revocation procedures;
staff and educator completion of role-specific training;
accessibility, privacy, quality, and incident findings closed within district targets;
number and type of exceptions, with reasons and expiry dates.

These measures show program coverage and follow-through. They do not prove that an AI system is error-free, unbiased, secure, or legally compliant.

Frequently asked questions

What is K-12 AI governance?

K-12 AI governance is the district operating model for deciding which AI uses are allowed, who may use them, for what purpose, with which data and model deployment, under what safeguards, and with what review evidence.

Is an AI policy enough?

No. A policy states intent. Governance also requires decision owners, an inventory, approval criteria, technical and procedural controls, training, evidence, incident handling, and periodic review.

Who should own district AI governance?

Executive accountability should be clear, but the work is cross-functional. Technology, curriculum, privacy, legal, security, accessibility, procurement, special education, school leadership, educators, and families each provide essential context.

How often should an approved AI use be reviewed?

Set a regular review date and trigger an earlier review after material changes to purpose, users, data, provider terms, model deployment, safeguards, or observed outcomes.

Does FERPA or COPPA automatically approve an AI tool?

No. Federal privacy rules provide requirements and possible legal models, not a universal product approval. Districts must evaluate the specific use, provider, data flow, contract, age group, and applicable state and local requirements.

Sources

For a plain-language application of this operating model across named student-facing products, read how school districts can govern student use of ChatGPT, Gemini, Claude, and other AI tools.

This guide is educational information, not legal advice. Districts should use qualified legal, privacy, accessibility, security, procurement, and instructional review for their circumstances.