Artificial intelligence in schools is entering a new phase.
For the past several years, much of the K–12 conversation has centered on whether teachers and students should use generative AI at all. Now, education agencies are beginning to confront a more difficult operational question: Which uses of AI are appropriate, which require additional safeguards, and which decisions should never be delegated to an algorithm?
The District of Columbia’s Office of the State Superintendent of Education, or OSSE, offered one of the clearest answers yet on September 1, releasing its first AI Model Policy for Staff Use for the 2026–27 school year.
The model policy gives local education agencies a practical framework for deciding how teachers, administrators and other employees may use artificial intelligence. At its center is a straightforward red-yellow-green “stoplight” system distinguishing prohibited applications from limited uses and lower-risk uses that remain subject to professional oversight.
The framework is guidance rather than law, and individual local education agencies may adapt it to their own circumstances. But its significance extends well beyond Washington.
D.C. is demonstrating what the next stage of institutional AI adoption may look like: not another list of recommended tools, but rules governing human judgment, student privacy, special education, assessment, teacher evaluation, cybersecurity and professional responsibility.
Read OSSE’s AI Model Policy announcement
Key Takeaways
- D.C. released its first model policy governing staff use of artificial intelligence in schools on September 1.
- The policy uses a red-yellow-green framework to distinguish prohibited, restricted and generally permitted uses.
- AI may not be used to make high-stakes decisions involving areas such as student discipline, teacher evaluation or eligibility for IEPs and Section 504 accommodations.
- More limited AI assistance may be used for functions including reviewing student work and drafting IEP language, but only with additional safeguards and human oversight.
- Lower-risk uses include lesson planning, instructional-material development, data analysis, communications and some administrative functions.
- OSSE recommends that employees use approved enterprise AI systems, receive AI-literacy training and renew that training annually.
- The policy was created after a February 2026 survey found that only 45% of D.C. local education agencies had established staff AI policies.
Schools Are Moving From AI Adoption to AI Governance
The change taking place in education is subtle but important.
The first generation of school AI policy often focused on students.
Can students use ChatGPT?
Does AI-generated writing constitute plagiarism?
Should generative AI websites be blocked?
What belongs in an academic-integrity policy?
Those remain important questions, but school employees now have access to AI systems capable of doing considerably more than generating a paragraph or brainstorming a lesson.
AI can summarize student records, analyze data, draft communications, evaluate writing, generate accommodations, suggest interventions and assist with administrative decisions.
That means school AI policy increasingly has to address not merely whether AI is used, but what authority schools are willing to give it.
OSSE’s policy draws that line around professional judgment.
State Superintendent of Education Dr. Antoinette S. Mitchell said the goal is to provide practical guardrails while keeping human judgment at the center of decision-making and strengthening privacy protections.
That principle may become one of the defining ideas in K–12 AI governance.
Artificial intelligence can assist educators.
It should not quietly become the decision-maker.
The Red Zone: Where AI Should Be Off-Limits
OSSE’s red category contains some of the policy’s most consequential provisions.
The model prohibits AI use for high-stakes decisions that depend on human judgment, including:
- physical surveillance of students or employees;
- student disciplinary decisions;
- teacher-performance evaluations;
- determining eligibility for individualized education programs;
- determining eligibility for Section 504 accommodations.
These restrictions matter because they address a temptation likely to grow as AI systems improve.
Many difficult school decisions involve large amounts of information.
AI systems can process information quickly.
That does not mean they should possess decision-making authority.
Consider student discipline.
An automated system might analyze behavior reports, attendance data or previous incidents. But disciplinary decisions may also require context about disability, school climate, individual circumstances, credibility and proportionality.
Teacher evaluation raises similar concerns.
Artificial intelligence could identify patterns in classroom data, but evaluating an educator involves professional observations and contextual judgments that cannot be reduced safely to an automated score.
The policy effectively separates analytical assistance from institutional authority.
That is a distinction other districts may want to consider carefully.
The Yellow Zone: AI Can Help, but Only With Safeguards
The yellow category addresses tasks where AI could provide real value but where mistakes could materially affect students or educators.
Examples include:
- monitoring activity on district-issued devices;
- drafting IEP language;
- reviewing or grading student work;
- providing supplemental coaching to educators.
This may be the policy’s most useful category because many real-world AI applications do not fit neatly into “allowed” or “prohibited.”
Take IEP documentation.
AI might help a special-education professional organize notes or draft language more efficiently.
But the student’s educational needs cannot be determined by an AI system.
Similarly, AI may help teachers review assignments, identify common misconceptions or draft preliminary feedback.
But an educator remains responsible for ensuring that the resulting judgment is accurate and appropriate.
The yellow category therefore recognizes something schools will increasingly encounter:
AI risk depends not merely on the tool, but on the consequence of the task.
Generating a first draft of a parent newsletter and helping evaluate a student’s academic performance may involve the same AI model.
They do not involve the same level of risk.
The Green Zone: Routine Professional Assistance
OSSE places lower-risk applications in its green category.
These include using AI to help:
- draft lesson plans;
- customize student-facing instructional materials;
- develop tutoring plans;
- analyze data sets;
- prepare school communications;
- support logistical operations.
Even these applications are not presented as autonomous.
OSSE’s framework still expects professional awareness and a human in the loop.
That matters for something as routine as lesson planning.
An AI-generated lesson can contain factual errors.
It can misinterpret standards.
It can propose activities inappropriate for a particular age group.
It can generate examples containing bias or invented information.
The productivity benefit may be substantial, but the educator is still the editor, subject-matter expert and responsible professional.
That principle could help districts move beyond the unproductive binary of “AI is dangerous” versus “AI saves teachers time.”
Both statements can be true depending on how the technology is used.
Only Approved AI Tools Should Handle School Information
One of the most operationally important elements of the policy concerns data.
OSSE recommends that school employees use approved enterprise tools, particularly when working with information connected to students.
For tasks involving personally identifiable information, AI systems must comply with applicable privacy and education requirements, including FERPA, COPPA, CIPA, IDEA, HIPAA where applicable, and the District’s own student digital-privacy law.
This addresses a growing challenge for districts.
A teacher may encounter dozens of consumer AI applications online.
Many are easy to access.
Some are free.
Few employees will independently understand each service’s data-retention practices, security model, contractual protections or treatment of uploaded information.
A seemingly harmless prompt such as:
“Help me create an intervention plan based on these student records”
could become a significant privacy issue if confidential information is entered into a tool the district has never approved.
AI governance therefore increasingly intersects with cybersecurity and procurement.
Districts need to know:
- which tools employees are using;
- what information those tools receive;
- whether prompts and uploads are retained;
- whether data may be used for model training;
- where data is stored;
- what contractual protections exist;
- who can access organizational accounts;
- what happens when an employee leaves.
The challenge is no longer simply educational technology.
It is enterprise technology governance.
Why D.C. Created the Policy
The policy did not emerge in a vacuum.
OSSE surveyed local education agency leaders in February 2026 about AI use and policy needs.
Only 45% reported having an established staff AI policy.
Respondents identified a state-level model policy as the most useful form of support OSSE could provide.
That figure illustrates the speed problem confronting school systems.
Teachers have already begun using AI.
Administrative systems increasingly contain AI functionality.
Education software vendors continue adding generative and predictive features.
But institutional policy has often developed more slowly than the technology.
The result can be a governance gap in which employees are experimenting with increasingly powerful systems while expectations differ from building to building—or even classroom to classroom.
A model policy reduces some of that burden.
Instead of expecting every district to develop a complete AI-governance framework independently, a state education agency can establish common principles that local leaders adapt.
That could become an increasingly important function for state departments of education.
AI Literacy Is Becoming a Staff Competency
OSSE’s framework does more than establish rules.
It also addresses training.
The agency recommends that school employees receive robust preparation before using AI, demonstrate AI literacy and renew their training annually. OSSE says it is adding two professional-development courses for D.C. educators: AI Literacy for Educators and Instructional Decision-Making and AI Dilemmas.
This is another important shift.
Schools have frequently treated AI knowledge as optional professional development for technology-interested teachers.
Policies like D.C.’s suggest that AI literacy could increasingly become an institutional competency.
That makes sense if educators are expected to make decisions about:
- hallucinations;
- bias;
- data privacy;
- appropriate disclosure;
- academic integrity;
- copyright;
- student information;
- AI-generated feedback;
- automated analysis.
Employees cannot follow an AI policy they do not understand.
TechEd has previously explored the growing role of AI professional development in AI Tools and Training in the Classroom, where teacher preparation emerged as a critical condition for effective adoption.
Recommended Reading: AI Tools and Training in the Classroom
Professional Judgment Becomes More Important, Not Less
AI is frequently described as a labor-saving technology.
In schools, however, widespread AI use may actually increase the importance of certain forms of professional judgment.
The educator becomes responsible for determining:
- whether AI should be used at all;
- whether the selected tool is approved;
- what information may be entered;
- whether the output is accurate;
- whether bias is present;
- whether the recommendation is appropriate;
- whether human review is sufficient.
That is a more complex role than simply accepting or rejecting artificial intelligence.
Good AI policy therefore cannot consist only of restrictions.
It has to teach employees how to reason about AI-assisted decisions.
D.C.’s stoplight system offers one practical method.
Before using AI, an educator can ask:
Is this a low-consequence productivity task?
Is this an activity involving meaningful student or employee consequences?
Is this fundamentally a decision that a human professional must make?
Those questions can travel with educators even as individual AI products change.
Special Education Deserves Particular Attention
The distinction between assistance and authority becomes especially important in special education.
OSSE places determining IEP or Section 504 eligibility in the prohibited red category while allowing AI assistance with drafting IEP language under the more restrictive yellow category.
That separation is instructive.
AI might reduce administrative burden by helping staff organize or draft documentation.
But eligibility decisions involve legal rights and individualized professional judgments.
The same principle could apply elsewhere.
An AI system might summarize assessment information.
It should not decide whether a student qualifies for services.
It might suggest possible accommodations.
It should not determine which accommodations a student receives.
It might help prepare meeting notes.
It cannot replace the IEP team.
Schools adopting AI should therefore examine workflows rather than simply tools.
The appropriate question is often not:
Can this AI system perform this task?
It is:
Which part of this task is appropriate for AI assistance, and which part must remain human?
AI Policy and Screen-Time Policy Are Beginning to Overlap
D.C.’s AI policy arrives as school systems are simultaneously reconsidering broader technology use.
TechEd recently examined this shift in School Screen-Time Policies Move Beyond Cellphone Bans, which documented how districts are beginning to scrutinize school-issued devices, streaming media, educational apps and AI functions—not just student-owned phones.
Recommended Reading: School Screen-Time Policies Move Beyond Cellphone Bans
These debates increasingly overlap.
A district can ban phones while expanding AI-enabled Chromebook use.
It can reduce social-media exposure while adding automated tutoring.
It can limit passive digital worksheets while encouraging sophisticated AI-supported research.
The important distinction is therefore becoming less about whether technology is present and more about what educational purpose it serves and what risks accompany it.
OSSE itself is addressing both issues. Its Education and Digital Technology initiative includes AI policy work as well as guidance related to personal wireless communication devices.
A Model Policy Is Not an AI Strategy
There is one important limitation.
D.C.’s new policy governs staff use.
It does not establish rules for student AI use, and it does not function as an AI-procurement policy. OSSE explicitly says those issues remain outside the scope of this document.
Districts therefore still need answers to broader questions.
What AI literacy should students possess?
When may students use generative AI on assignments?
How should schools disclose automated systems?
How will districts evaluate AI vendors?
Who maintains an inventory of AI-enabled applications?
How are parents informed?
What outcomes should AI actually improve?
A mature AI strategy will ultimately need to connect all of those areas.
The staff policy is one layer.
It is an important one.
What District Leaders Should Do Now
School systems do not need to copy D.C.’s policy word for word.
They should consider conducting the same exercise.
Start by identifying actual uses of AI across the organization.
Do not assume leadership already knows them.
Teachers may be using AI for lessons and feedback.
Counselors may be experimenting with AI to draft communications.
Administrators may be analyzing spreadsheets.
Human-resources staff may be using AI to rewrite job descriptions.
Technology departments may have purchased platforms that recently added AI functionality.
Once those activities are identified, classify them by risk.
A stoplight system is attractive because employees can understand it.
The categories can then be supported by more detailed policy language covering privacy, data governance, cybersecurity, procurement and professional accountability.
Finally, provide training.
A policy posted on an intranet is not implementation.
Questions to Ask Your District
- Do we have a written policy governing staff use of generative AI?
- Do employees know which AI tools are approved?
- Can staff identify information that should never be entered into consumer AI systems?
- Have we separated AI assistance from decisions requiring human professional judgment?
- Which high-stakes uses of AI are explicitly prohibited?
- Do our rules address special education and Section 504 processes?
- Can teachers use AI when reviewing or grading student work?
- What level of human review is required?
- Do employees receive formal AI-literacy training?
- How often should that training be renewed?
- Do we maintain an inventory of AI-enabled platforms?
- How are new AI features evaluated when existing vendors introduce them?
- Does our cybersecurity team participate in AI procurement?
- How do we verify compliance with student-data privacy requirements?
- Are student-use policies consistent with staff-use policies?
- Who is accountable when an AI-generated recommendation is wrong?
What Educators Should Watch Next
Student AI Policies
The next major policy frontier is likely to be student use.
Districts still need workable rules that distinguish AI-assisted learning from academic substitution while recognizing that blanket prohibitions become increasingly difficult as AI is embedded in common software.
AI Procurement
Schools will need procurement frameworks specifically designed for artificial intelligence.
Traditional software reviews may not adequately address model training, prompt retention, automated decision-making and rapidly changing product functionality.
Special Education
The distinction between AI-assisted documentation and AI-driven eligibility or accommodation decisions is likely to receive increasing attention.
Teacher Evaluation
As analytics become more sophisticated, districts may face pressure to incorporate automated analysis into employee evaluation.
D.C.’s decision to place teacher-performance decisions firmly in the prohibited category provides an important policy precedent.
State-Level Guidance
Only 45% of the D.C. LEAs surveyed had established staff AI policies before OSSE developed its model.
Other states may encounter similar gaps.
Model state frameworks could become a common way to prevent dozens or hundreds of districts from independently solving the same governance problem.
Frequently Asked Questions
What is D.C.’s new school AI policy?
The Office of the State Superintendent of Education released an AI Model Policy for Staff Use on September 1, 2026. It provides voluntary guidance that D.C. local education agencies can customize when developing their own rules for employee use of artificial intelligence.
Is the policy mandatory?
No. OSSE describes it as customizable guidance rather than legal advice. LEAs may adopt or modify it according to local needs.
What is the stoplight framework?
The framework divides AI applications into red, yellow and green categories. Red uses are prohibited, yellow uses require additional safeguards, and green uses are generally permitted but still require professional awareness and human review.
What uses of AI are prohibited?
Examples include physical surveillance, student disciplinary decisions, teacher-performance evaluations and determining eligibility for IEPs or Section 504 accommodations.
Can teachers use AI to grade student work?
OSSE places reviewing and grading student work in its yellow category, meaning limited use may be appropriate with safeguards and human oversight.
Can teachers use AI to write lesson plans?
Yes. Lesson-plan drafting is among the lower-risk uses identified in the green category, although educators remain responsible for reviewing the resulting material.
Can AI be used for IEPs?
OSSE distinguishes between tasks. AI may provide limited assistance with drafting IEP language under appropriate safeguards, but determining whether a student qualifies for an IEP or Section 504 accommodation falls into the prohibited category.
Does the policy cover student use of AI?
No. This particular model policy focuses on staff use and does not establish student AI rules.
Does the policy address data privacy?
Yes. OSSE recommends approved enterprise AI tools and emphasizes compliance with applicable privacy requirements whenever personally identifiable information is involved.
Does D.C. require AI training for teachers?
The model recommends robust training before staff use AI, demonstrated AI literacy and annual renewal of training. OSSE also plans additional AI professional-development courses for educators.
TechEd Magazine Perspective
The most important contribution of D.C.’s new policy may be its simplicity.
Artificial intelligence is complicated.
School policy does not always need to be.
The red-yellow-green structure gives educators a question they can apply in ordinary work:
How much consequence are we giving this machine?
Using AI to draft a newsletter is fundamentally different from using it to evaluate a teacher.
Generating a lesson outline is fundamentally different from determining whether a child qualifies for special-education services.
Analyzing a spreadsheet is fundamentally different from deciding whether a student should be disciplined.
Good AI governance recognizes those distinctions.
School systems that attempt to regulate AI solely through a list of approved websites will struggle because the products will keep changing.
AI capabilities will appear inside tools districts already use.
New models will arrive.
Features will change.
What can remain stable are the principles underneath the policy:
Protect student information.
Match safeguards to risk.
Require professional review.
Do not outsource consequential human decisions.
Train employees to understand the technology they are using.
D.C.’s framework is unlikely to be the final answer to K–12 artificial-intelligence governance.
It may nevertheless offer districts a useful way to begin asking the right questions.
Recommended Reading
Using AI in Schools
TechEd’s earlier examination of how schools can move beyond blanket AI bans toward purposeful instructional use and responsible implementation.
Read Using AI in Schools
AI Tools and Training in the Classroom
Explores educator AI adoption and why professional development is becoming essential as generative AI enters routine classroom practice.
Read AI Tools and Training in the Classroom
School Screen-Time Policies Move Beyond Cellphone Bans
Examines how districts are beginning to govern school-issued technology, educational applications and AI features as part of the larger digital-learning environment.
Read School Screen-Time Policies Move Beyond Cellphone Bans
Technical and STEM Education 2026
TechEd’s broader look at the changing landscape of STEM, technical education and workforce preparation.
Read Technical and STEM Education 2026
Sources
D.C. Office of the State Superintendent of Education — AI Model Policy Announcement
Primary source for the September 1 policy release, the stoplight framework, survey findings, privacy requirements and planned educator training.
Read the OSSE announcement
OSSE — Education and Digital Technology
Official resource hub for D.C.’s artificial-intelligence policy work, digital-technology guidance and related resources.
Visit OSSE Education and Digital Technology
OSSE — 2026–2030 Strategic Plan Overview
Provides broader context for D.C.’s responsible-AI work, including privacy, equity, transparency, staff preparation and responsible AI in teaching and school operations.
Read OSSE’s strategic-plan overview
