Implementation Guide
IEP Documentation AI Governance: Runtime Controls for FERPA and IDEA Compliance
IEP documentation AI governance requires applying FERPA's school-official exception and IDEA's individualized, team-based decision requirement to AI agents through runtime controls: least-privilege access scoped to a single student and task, tool-call allowlisting against the SIS, immutable audit logs of every data access, and mandatory human review before AI-drafted content becomes part of the official IEP record.
In short: Districts satisfy FERPA and IDEA obligations for AI agents in IEP workflows by enforcing four runtime controls at the technical layer: per-student access scoping, tool-call allowlisting against the SIS, immutable audit logging, and mandatory human review before any AI-drafted content is finalized.
Least-Privilege Access Architecture for AI Agents Touching Student Records
These four controls define how an AI agent's access to student data should be constrained at runtime, independent of the policies a vendor states in a contract.
- 1
Per-student scoping
Access is limited to the specific student record and IEP task in progress, not database-level or district-wide SIS access.
- 2
Read/draft vs. write/finalize separation
Actions that generate draft language are architecturally distinct from actions that commit changes to the official IEP record.
- 3
Tool-call allowlisting
Runtime policy restricts which SIS functions, data fields, and document types an agent can invoke, rejecting out-of-scope calls by default.
- 4
Context-layer data minimization
Prompt and context construction exclude unrelated students' records or unrelated disability categories during multi-step agent tasks.
Evaluation Questions for AI Governance in IEP Systems
Use these questions during procurement or renewal to assess whether a vendor's AI system meets baseline governance requirements before it touches student data.
- Does the vendor contract satisfy FERPA's school-official exception, including a direct-control clause and prohibition on data reuse or model training?
- What runtime controls limit the AI agent's access to only the specific student record and IEP task authorized, versus broader SIS access?
- What audit log fields are captured for AI agent actions on IEP data, and how long are logs retained?
- Is there a mandatory human review and approval step before AI-drafted IEP content is finalized?
- How does the system prevent or flag AI-generated content that could be construed as an automated eligibility or service-level determination?
FERPA and IDEA Compliance Requirements for AI in IEP Workflows
FERPA (34 CFR Part 99) restricts disclosure of personally identifiable information from education records unless an exception applies. The most relevant exception for AI vendors is the school-official exception, which permits data sharing without parental consent only if the district retains direct control over how the vendor uses and maintains the data, and the vendor does not redisclose or repurpose it, including for model training. This makes vendor contract language a compliance-critical technical requirement, not just a legal formality.
IDEA imposes a separate obligation: an IEP must be developed by a team that includes the parent and qualified professionals, and the resulting document must reflect individualized determinations about a child's needs and services. Neither statute contains AI-specific provisions. Districts are left to apply existing disclosure, consent, and individualized-decision requirements to systems that did not exist when the regulations were written. That gap does not remove the obligation; it shifts the burden onto districts to demonstrate, through documented technical controls, that an AI system's behavior satisfies these existing rules.
Audit Logging and Human-in-the-Loop Controls
FERPA's recordkeeping provisions require districts to maintain records of certain disclosures of personally identifiable information. Applied to AI agents, this implies that any agent action touching protected student data should generate an auditable record: which user initiated the task, what data the agent accessed, what the agent produced, and when the action occurred. Without this trail, a district cannot readily demonstrate compliance during a records request or due process proceeding.
IDEA's team-based decision requirement has a parallel implication for workflow design. AI-generated language, whether a draft goal, a summary of present levels of performance, or proposed service minutes, should be treated as unfinished input to the IEP team's judgment, not as content that enters the official record without review. A mandatory human sign-off step before AI-assisted content is finalized keeps the legal decision-maker where IDEA requires it to be: with the IEP team, not the model.
Scope and Limitations of Current Guidance
No federal regulation or Department of Education rule currently addresses AI agents in IEP documentation specifically. The 2023 Office of Educational Technology report on AI in teaching and learning offers general recommendations on human oversight and risk evaluation, but it is non-binding guidance, not enforceable regulation, and it is not IEP-specific. NIST's AI RMF (2023) provides a voluntary structure useful for organizing governance documentation during procurement or audit, but it likewise contains no IEP-specific technical standard.
State student privacy statutes vary and may impose requirements stricter than FERPA; districts should confirm state-level obligations before deploying AI tools district-wide. No due process hearing precedent addressing AI-assisted IEP content was identified at the time of this guide. Districts should treat the controls described here as a reasonable, defensible interpretation of existing law, not as confirmation from a regulator or court that any specific implementation is compliant.
Core Obligations AI Agents Must Satisfy
Four legal requirements, translated into runtime technical controls, form the foundation of a defensible AI governance posture for IEP workflows.
FERPA School-Official Exception
AI vendors may access IEP data without consent only under direct district control, with no redisclosure or secondary use.
IDEA Team-Based Decision Requirement
IEP content must reflect individualized determinations made by a team, not autonomous AI output.
FERPA Recordkeeping Obligation
Districts must maintain records of certain disclosures of personally identifiable information under 34 CFR 99.32.
No AI-Specific Federal Rule
Districts must translate existing FERPA and IDEA obligations into runtime technical controls themselves.
Apply Runtime Governance to AI Agents in IEP Workflows
Least-privilege access, tool-call restrictions, and audit logging are technical controls, not policy statements. Districts evaluating AI tools for IEP documentation should confirm these controls are enforced at runtime before deployment.
Explore Runtime Governance