AI in Education

AI in Education explains how artificial intelligence can support teaching, learning, assessment, administration and research while creating new risks around accuracy, privacy, bias, academic integrity and overreliance.

Educational institutions remain accountable for decisions made with AI. A model can suggest, summarize or generate, but educators and institutions must define the learning purpose, verify outputs and protect learners.

On this page — 19 sections

1. AI-in-education governance framework
2. Common AI use cases
3. Human-centered use
4. AI literacy
5. Accuracy and hallucination
6. Assessment and academic integrity
7. Privacy
8. Bias and fairness
9. Accessibility
10. Teacher planning
11. AI tutoring
12. Procurement
13. Pilot design
14. Policy contents
15. Common failures
16. AI incident response
17. Frequently asked questions
18. Related Knowledge Compass guides
19. Official sources and review notes

AI-in-education governance framework

Function	Education question
Govern	Who is accountable for the use case?
Map	Who is affected and what can go wrong?
Measure	How are accuracy, bias and learning value tested?
Manage	Which controls, review and escalation are required?
Review	When should the system be changed, limited or stopped?

This structure follows the NIST AI Risk Management Framework functions and adapts them to educational use.

Common AI use cases

Use case	Potential value	Primary risk
Tutoring	Immediate explanation and practice	Incorrect or overconfident guidance.
Teacher planning	Draft activities and examples	Weak alignment and hidden errors.
Feedback	Faster formative comments	Generic, biased or invalid judgments.
Accessibility	Alternative formats and language support	Inaccurate captions or simplification.
Advising	Information and reminders	Consequential wrong advice.
Assessment design	Question and rubric drafts	Leakage and invalid items.
Research support	Search, coding and synthesis	Fabricated sources and privacy.
Administration	Classification and communication	Incorrect records or exclusion.
Analytics	Patterns and risk flags	Surveillance and false prediction.

Human-centered use

AI should strengthen educator judgment and learner agency rather than replace them. The institution should state which outputs require review and which decisions cannot be delegated.

• Teachers remain responsible for instructional decisions.
• Learners should know when AI is used.
• High-consequence advice requires a human route.
• Generated content needs an accountable owner.
• Students need opportunities to question outputs.
• Accessibility and non-AI alternatives should remain available.

AI literacy

Audience	Capability
Students	Prompting, verification, citation, privacy and limits.
Teachers	Learning design, review, bias and assessment redesign.
Leaders	Governance, procurement, risk and evidence.
IT and security	Access, logging, integration and incident response.
Support staff	Approved use, escalation and student communication.
Parents or guardians	Purpose, data use and available choices.

Accuracy and hallucination

Generative systems can produce fluent but incorrect information, fabricated references or inconsistent reasoning. Educational use requires source checking and a design that does not reward confident presentation over evidence.

1. Use approved source material.
2. Request citations.
3. Open and verify each source.
4. Check calculations and examples.
5. Test edge cases.
6. Record known limitations.
7. Provide a correction process.

Assessment and academic integrity

AI changes what unaided work can demonstrate. Institutions should clarify permitted assistance, teach disclosure and redesign assessment around process, oral explanation, practical application and source use.

Assessment approach	Integrity benefit
Draft history	Shows development and revision.
Oral defense	Tests understanding and decisions.
Personal or local data	Reduces generic answer usefulness.
Practical task	Shows performance.
In-class checkpoint	Provides comparison evidence.
Reflection	Explains tool use and verification.
Multiple evidence sources	Avoids one fragile signal.

Detection scores should not be treated as definitive proof. Investigations need due process, context and human review.

Privacy

Prompts may contain student work, disability information, behavior, personal data or unpublished research. Institutions should define approved tools and prohibit sensitive data in systems without appropriate agreements and controls.

• Data minimization.
• Role-based access.
• Retention and deletion.
• Subprocessor review.
• Training-use settings.
• International transfer review.
• Student notice.
• Incident response.

Bias and fairness

AI performance can vary across language, disability, culture, dialect and subject. Institutions should test representative users and avoid using unvalidated outputs for admission, grading, discipline or placement.

Accessibility

AI can produce captions, descriptions, summaries and translation, but generated accessibility support requires review. The underlying tool and workflow should be tested against current accessibility standards and with assistive technology.

Teacher planning

AI can accelerate first drafts of lesson plans, examples and questions. Educators should verify curriculum alignment, reading level, cultural context, factual accuracy, accessibility and cognitive demand.

AI tutoring

A tutor should be bounded to an approved subject, identify uncertainty and avoid high-stakes advice. It should encourage learner reasoning rather than simply produce final answers.

Tutor control	Purpose
Source grounding	Reduce unsupported answers.
Hint progression	Support productive struggle.
Refusal boundaries	Avoid unsafe or out-of-scope advice.
Teacher visibility	Support review without excessive surveillance.
Escalation	Route confusion or risk to a person.
Learning evidence	Measure understanding beyond chat volume.

Procurement

1. Define one educational use case.
2. Identify affected learners.
3. Review model and data flow.
4. Test representative content.
5. Check accessibility.
6. Review privacy and security terms.
7. Define human review.
8. Plan export and deletion.
9. Measure educational value.
10. Set renewal criteria.

Pilot design

Begin with a low- or moderate-risk use case and a small, representative group. Compare the AI-supported process with the current approach.

• Baseline time and quality.
• Accuracy after review.
• Learner understanding.
• Teacher workload.
• Access across groups.
• Privacy or safety incidents.
• Override and escalation rates.
• Cost per useful outcome.

Policy contents

• Approved and prohibited tools.
• Permitted student assistance.
• Disclosure and citation.
• Sensitive-data rules.
• Assessment expectations.
• Human review.
• Accessibility and accommodations.
• Incident reporting.
• Research and intellectual property.
• Policy review date.

Common failures

• Buying an AI feature before defining the need.
• Treating output fluency as accuracy.
• Using student data for unrelated purposes.
• Automating consequential decisions.
• Requiring AI without an alternative.
• Using detection as proof.
• Failing to train teachers and students.
• Measuring chat volume instead of learning.
• Ignoring model changes.
• Keeping a tool after evidence is weak.

AI in education should expand human capability and learner agency without weakening evidence, privacy or accountability.

AI incident response

1. Capture the prompt, output and affected decision.
2. Protect student and staff data.
3. Pause high-risk use.
4. Notify academic, privacy, security and legal owners.
5. Correct affected content or decisions.
6. Check for similar outputs.
7. Update policy and controls.
8. Document closure.

Students and educators need a simple way to report harmful, inaccurate or discriminatory output without penalty.

For individual credential pathways, compare AI Certifications Worth Getting.

Frequently asked questions

Related Knowledge Compass guides

• What Is EdTech?
• Adaptive Learning Platforms Explained
• Online Exam and Proctoring Tools
• AI in Corporate Learning
• Education Technology & EdTech

Official sources and review notes

Last reviewed: June 2026. Models, policies and legal requirements change quickly.

• NIST AI Risk Management Framework
• NIST Generative AI Profile
• UNESCO Technology in Education
• U.S. Department of Education Student Privacy
• W3C WCAG 2.2

↑ Back to top