What you’ll learn in this article…
- FERPA and COPPA can trigger obligations from one unapproved student upload.
- Vendors that refuse to sign a data processing agreement are red flags.
- The PowerSchool breach exposed about 62 million student and teacher records.
PowerSchool's late-2024 breach exposed records for about 62 million students and teachers. Generative AI sharpens that risk: one student photo uploaded for quick feedback can trigger FERPA and COPPA obligations.
That is not hypothetical. A 2025 Reddit thread shows teachers uploading student photos to generative AI for parent notes and rubrics, often before checking district policy. The tools feel ordinary; the data governance questions are not.
District leaders are responding with stricter AI vetting, but teachers often lack clear guidance. The question is no longer whether to use AI, but how to use it without creating legal exposure.
Where AI Shows up in K-12 Classrooms: Which Data Flows Create Risk
OpenAI's ChatGPT Team and Edu plans retain student and staff prompts indefinitely by default, with a 90-day minimum retention window as of 2025.1 That single default setting explains why teachers need to map AI use before uploading anything.
Common Uses and the Data They Touch
- Writing assistants and feedback tools: student essays, names, assignment prompts, and sometimes details from IEPs or behavior plans.
- Reading and adaptive tutoring platforms: assessment scores, reading level, time on task, and error patterns.
- AI image generators: student photos, artwork, or likeness used to create or edit images.
- Generative chatbots for AI lesson planning: snippets of student work or class discussion notes pasted into a third-party prompt.
Among these, photos and IEP details are highest sensitivity because they identify a minor and reveal disability or behavioral status. Essays and assessment scores come next; usage logs and anonymous practice data are lower but still linked to a student account.
Where the Data Goes After You Upload
Once data leaves a school-controlled account, teachers rarely see the full lifecycle. ChatGPT workspace training is opt-in, but retention can still be indefinite.1 Google Gemini for Education states student data is not used to train models, not human-reviewed, not shared outside the school domain, and not used for advertising.2 Khanmigo Lite sends messages to OpenAI, which may use them to improve models.3 IXL may use de-identified data for service improvement.4 Free tools often fund operations through data monetization, and K-12 students cannot consent for themselves, which makes ethical AI use defaults more consequential than in higher ed.
Student Data Privacy Laws Every Teacher Should Understand
Using AI tools with students forces a choice between instructional convenience and legal exposure. A single unapproved upload can trigger obligations under federal and state law, even when no one intended harm.
FERPA: The Baseline for Education Records
FERPA protects education records. It permits a school to share those records with an outside AI vendor only when the vendor qualifies as a "school official" with a "legitimate educational interest." That status is not automatic.1 The district must have a written contract that keeps the vendor under direct control, limits use to the specific educational service, and forbids redisclosure.1 General AI tools qualify only if the district formally designates them as school officials and documents why each data element is necessary. AI vendors cannot use student data to train their models under this exception.
COPPA: Separate Rules for Under-13 Students
COPPA applies to students under 13. Operators must use COPPA verified parental consent methods before collecting personal data from children under 13. Schools can provide consent for educational uses only, not for AI training or third-party sharing. The COPPA 2025 amendments, which became effective June 23, 2025 with a compliance deadline of April 22, 2026, require separate parental consent for AI training. There is one narrow audio exception: a child's voice request can be processed without consent only if the audio is deleted immediately and not used for anything else.4 Most consumer AI tools are not built to meet these requirements.
State Laws Often Go Further
California SOPIPA prohibits targeted advertising, sale of student data, and profiling except for K-12 school purposes.1 New York Education Law 2-d requires written third-party contractor agreements with encryption, incident response, and no marketing use.1 Texas HB 18 adds heightened protections against profiling of minors.3 Teachers should assume the strictest applicable state law governs any AI use.
Your Personal Exposure Goes Beyond District Liability
Using an unapproved tool that causes a data breach can lead to professional discipline, including teacher certification suspension, not just district sanctions. If you sign up for a consumer AI tool on your own and upload student work, you may have bypassed required contract safeguards and consent workflows. In higher education, FERPA still applies, but COPPA does not. Students 18 and older can consent for themselves, shifting the governance model from parental consent to student decision making.
Key Questions to Ask Before Any AI Tool Reaches Your Students
Use this checklist before any AI tool reaches students. A vendor that will not sign a data processing agreement (DPA) is a red flag by itself, and legitimate edtech providers will sign one. Have district IT or legal counsel review any contract before tools access sensitive records such as student photos, IEPs, or behavioral data, and confirm the vendor can support any required parent or guardian consent workflow.
| Evaluation Dimension | What to Ask the Vendor | Green Flag | Red Flag |
|---|---|---|---|
| Model training use | Does your DPA explicitly prohibit using any student data, including submissions, responses, or behavioral data, for model training, fine-tuning, or product improvement without separate written authorization? | Vendor DPA states it will not use student submissions, responses, or behavioral data to train, fine-tune, or improve any model without separate written authorization. | AI tool uses student submissions to train models or improve products for broader commercial use without explicit institutional authorization and appropriate consent. |
| Data processing scope and purpose | Is the DPA clear that student data will only be processed for the specific educational services we contract for, and not for unrelated analytics, marketing, or sales? | DPA limits use of student data to providing the contracted educational service and bars use for any other purpose unless explicitly authorized with appropriate consent. | Policy language allows broad use of student data for general product development, analytics, or commercial purposes beyond the contracted educational service. |
| Third-party sharing and subprocessors | Can you provide a complete list of all subprocessors, including cloud providers, model providers, annotation services, and inference services that may access student data, and will you notify us before changes? | Vendor provides a complete subprocessor list for model infrastructure, including cloud compute providers, annotation services, and fine-tuning partners. | Vendor does not disclose all third-party subprocessors that may access student data or refuses to commit to advance notification when subprocessors change. |
| Data retention and deletion | What are your timelines and processes for returning or deleting all student data at contract termination, including backups and logs, and can you delete an individual student's data on request? | DPA specifies data return or deletion upon contract termination, including production systems, backups, and logs, and vendor confirms the ability to identify and delete an individual student's data from training sets and inference logs when requested. | Vendor cannot remove a specific student's data from models, training datasets, or logs, or says deletion is not technically possible. |
| Student PII in inputs and outputs | Will our use of the tool require student personally identifiable information (PII) as input, and could the tool's outputs also be considered student PII? | Vendor helps clearly determine when inputs and outputs constitute student PII and supports risk mitigation when substantive decision-making is involved. | Vendor cannot explain whether student PII is involved in inputs or outputs, or downplays risks for use cases involving substantive decisions about students. |
| Redisclosure risks from generative AI outputs | How do you ensure that student PII used by the tool will not be redisclosed or regenerated in outputs, either to the same user or other users? | Vendor has documented safeguards to prevent redisclosure of student PII in generative AI outputs and can explain how those controls work. | Vendor has no clear strategy to prevent redisclosure of student PII or dismisses the risk that the model could reproduce sensitive information in outputs. |
| Privacy policy transparency | Do your policies clearly disclose all key privacy and security practices in an educational context, and can you point to specific policy sections for each practice? | Vendor policies allow a clear overview, triage, and evaluation of privacy and security practices, including references to specific policy text for each issue. | Policies omit many of the privacy and security issues commonly expected to be disclosed for educational products or are too vague to evaluate. |
| Privacy best practices alignment | Can you confirm that your product aligns with key edtech privacy best practices such as no data sales, no third-party marketing, and no targeted ads to students? | Vendor adheres to Common Sense-style best practices: no data sales, no third-party marketing, no targeted advertisements, no third-party tracking, no cross-app tracking, and no commercial profiling. | Vendor engages in third-party marketing, targeted advertising, or commercial profiling of students, or sells or shares data for advertising uses. |
Related Articles
What to Do When You Are Asked to Upload Student Data to an AI Tool
Being asked to upload student data into an AI in the Classroom tool usually starts small: a principal shares a new platform, a vendor demo looks promising during professional development, and using it feels like the default next step. The pressure is real, but the decision to upload student information deserves a pause before it becomes routine.
Red Flags That Should Make You Stop
Even one of these signals is a reason to ask for written approval first:
- The tool was not approved by district IT or instructional technology.
- There is no data privacy agreement (DPA) on file between the district and the vendor.
- The free tier explicitly states that inputs may be used to train the model.
- The tool asks for student names, photos, or contact information without a clear instructional need.
Lower-Risk Ways to Proceed
If you still want to test the tool, choose safer routes before involving real student records:
- Use fabricated sample data first to see how the tool works.
- Replace names with student IDs, strip photos and contact fields, and use aggregated class data rather than individual records.
- Use only district-approved tools that already have DPAs and data handling terms.
Understand What De-identification Really Means
Under FERPA, the de-identification standard means removing all directly identifying information and making a reasonable determination that no reasonable person in the school community, without personal knowledge, could identify the student with reasonable certainty. That standard is not the same as deleting a name. Indirect identifiers matter: a student's grade level, school, disability status, and pattern of scores can combine to make someone re-identifiable even when a name is gone. This is sometimes called the mosaic effect. Removing direct identifiers alone is rarely enough when multiple small clues remain.
If you do de-identify, send only the minimum necessary data1 and document your reasoning2. There is no single required statistical method2, but the district should be able to explain why the remaining data is unlikely to identify a student.
If You Are Pressured Anyway
If an administrator or vendor urges you to use an unapproved tool with real student data, document the request: the date, who asked, what data, and what purpose. Loop in your district privacy officer, data protection officer, or union representative. Ask for written approval before proceeding. A verbal okay is not enough when student data is involved.
Building a Classroom AI Policy and Consent Workflow That Actually Works
The hardest part of classroom AI policy is balancing instructional flexibility with legal consent requirements, especially when a promising new tool asks you to upload student photos, essays, or other personal data. A clear workflow can protect you and your students without making every tool feel like a legal hurdle.
COPPA Consent Is Not a One-Time Form
For students under 13, the Children's Online Privacy Protection Act (COPPA) requires verifiable parental consent before an operator collects personal information. A blanket technology use agreement signed at enrollment often does not satisfy this for a new AI tool, because that tool may collect different data types or use student inputs for AI training.1 If the operator uses data for noneducational purposes such as training its models or targeted advertising, COPPA requires separate parental consent beyond the school-authorization framework.2 Districts commonly use an opt-out model for strictly educational tools1, but noneducational AI uses require opt-in2.
Building an Opt-In Workflow for Higher-Risk AI
For AI tools that process photos, essays, or other sensitive student work, treat consent as opt-in: the default is off, and active parent consent is required before any student data flows.2 Lower-risk, district-approved tools may use opt-out, where parents must act to remove a student.1 Use this workflow:
- Notify parents: Identify the specific tool, the personal information it collects, retention practices, third-party sharing, and whether data is used for AI training.
- Offer a real opt-out: Provide an alternative that does not penalize the student academically. If a parent declines, the student should receive equivalent instruction without the tool.5
- Document responses: Record the consent method, date, and consenting adult. Keep records for the entire school year.
- Use layered consent: Separate checkboxes for educational use and noneducational uses like AI training or biometric capture, modeled on a COPPA Parent/Guardian Consent Form.3
A Simple Classroom Policy Structure
Post a one-page policy that lists approved tools, prohibited data types, and the reason behind the rules. For example, prohibit student photos in consumer AI apps and IEP content in unapproved tools. Explain the policy in student-friendly language, such as: We protect your information so we can use AI safely. Translate parent notices into languages families actually read. A COPPA notice full of legalese does not satisfy the spirit of informed consent.1
Teaching Students to Understand Their Own Data Rights
Students can either treat AI in the Classroom tools as invisible helpers and surrender their information by default, or they can learn to read how those tools collect, store, and use data. Treating student data literacy as an extension of media literacy prepares them to protect themselves long after they leave your classroom.
Age-Differentiated Discussion Starters
- Elementary: Frame privacy in terms of private vs public information. Ask students why they should not share a full name, address, or birthday online, and whose job it is to keep that information safe.
- Middle school: Show how apps collect data through permissions, cookies, and sign-in screens. Discuss what companies might do with that data, from targeted ads to building a profile over time.
- High school: Examine how AI models are trained on user data and what "data as payment" means. Ask students what a free tool is actually costing them when it asks for photos, voice recordings, or writing samples.
Free Resources and a Classroom Activity
Common Sense Media's Digital Citizenship curriculum and the iKeepSafe program both offer free, age-appropriate lessons on privacy, identity, and data literacy, including teacher guides and family-facing materials.
For a concrete activity, have students open the privacy policy of a tool they already use, such as a learning app or social media platform. Ask them to identify three specific types of data the tool collects and write down how each one might be used. Then lead a discussion: Does this collection feel fair? What would you change if you could rewrite the policy? That shift from passive user to policy reader is where real data autonomy begins.
Professional Development and AI Literacy Resources for Educators
What professional development actually prepares teachers to evaluate AI tools for student privacy, and where should educators start?
Ongoing AI professional development for teachers is no longer optional enrichment. The AI tool landscape shifts faster than district policy can be updated, so teachers who stay current with AI in the classroom are the first line of defense for student data. Treat PD as a professional obligation tied to your duty of care, not a checkbox.
Start With Established Frameworks and Free Resources
- ISTE's AI in Education framework provides concrete competencies for evaluating and using AI responsibly.
- CoSN's Trusted Learning Environment program helps schools assess their data privacy practices.
- The U.S. Department of Education's 2023 report "Artificial Intelligence and the Future of Teaching and Learning" outlines ethical considerations and policy guidance.
- Common Sense Media offers free educator privacy tools and AI literacy lessons you can adapt.
Advocate for School-Level AI Governance
Professional development works best when paired with structure. Push for a standing AI review committee, a maintained approved-tools list, and a clear escalation path for teacher concerns about vendor data practices. These structures convert individual learning into institutional protection.
Distinguish K-12 from Higher Ed Needs
K-12 teachers carry COPPA obligations because they work with minors, and data collection often requires parental consent. Higher education faculty may have more flexibility in tool selection, but they still operate under FERPA and often institutional IRB requirements when using student data for research or classroom experiments. Tailor your PD accordingly.
Share Vetted Resources With Colleagues
Privacy literacy spreads faster through peer networks than top-down mandates. After you evaluate a tool or complete a training, share what you learned with your grade-level team or department. A shared resource folder or short staff meeting demo can multiply your impact.
There are two ways to bring AI into your classroom: avoid it and lose instructional value, or adopt it without checking data flows and risk legal exposure. The question is not whether to use AI in the Classroom; it is how to use it in a way that keeps student data safe and legal.
Start with one concrete step: identify a single tool you currently use with students and run it through the vendor vetting checklist. If the vendor will not sign a data processing agreement, do not put student data in it. Every data decision is also an ethical decision. Students trust you with their information; protect that trust before you hit upload.
