AI in Education Safety: Build Guardrails That Protect Learning and Student Well-Being
A chatbot can give a student a polished answer in seconds. That answer may also be wrong, unhelpful, or harmful to learning. This tension sits at the center of Erin Mote's testimony before the Senate HELP Committee. As CEO of Innovate EDU, a charter school founder, technologist, and parent, Mote argued that AI in education safety must come before broad classroom adoption.
Her central point is clear: safety creates the trust and structure that responsible innovation needs. Schools should judge AI by its effect on learning, student well-being, access, and human judgment.
Why AI in Education Safety Must Come Before Classroom Adoption
Generative AI Has Already Entered Classrooms
Mote called generative AI an "arrival technology," comparable in social impact to the internet or electricity. Her testimony cited figures showing that about 85% of teachers and 84% of students already use generative AI in classrooms.
Adoption moved faster than school procurement, privacy review, risk checks, and teacher training. Families and educators now face thousands of tools, unclear rules, and uneven guidance. The question is no longer whether students will meet AI. Schools must decide how students will use it safely.
Training Gaps Weaken Responsible Use
Mote said more than half of schools have provided no professional development on safe AI use. Without training, teachers may struggle to spot false answers, protect student data, or choose tools that support instruction.
A basic school AI plan should include staff training, approved-tool lists, privacy reviews, classroom rules, and a clear process for reporting harm. Federal funding and shared guidance could help districts build these systems instead of forcing each school to start alone.
Safety also supports innovation. Teachers are more likely to try useful tools when they understand the risks. Families are more likely to trust schools that explain how student data and AI systems are handled.
How Consumer Chatbots Can Undermine Student Learning
Engagement Is Not the Same as Teaching
Consumer language models often aim to keep users active and satisfied. Their friendly responses can validate a false idea, provide an answer too soon, or remove the effort needed to solve a problem.
That effort is called productive struggle. Students build reasoning and persistence when they work through confusion, test ideas, and revise mistakes. A purpose-built learning tool can offer hints, feedback, and practice without taking over the task.
Mote pointed to AI math tools used in Alabama as an example of targeted support for students with weak foundations. These systems focus on a defined skill instead of trying to answer every question.
Cognitive Surrender Can Reduce Critical Thinking
Mote cited Warren researchers who use the term "cognitive surrender" for the loss of independent judgment that can occur when AI agrees too easily. She also cited a finding that users accepted intentionally wrong chatbot answers 80% of the time. The study, sample, and test conditions need review before that figure is used as settled evidence.
Teachers can reduce this risk by making students critique AI output. Students should verify claims, explain their reasoning before seeing an AI response, and identify bias, uncertainty, and made-up information. For some neurologically vulnerable youth, these habits may be especially important.
How Evidence-Based EdTech Purchasing Can Reduce Risk
Thousands of Tools Create Unnecessary Work
Mote said the average district accesses nearly 3,000 digital tools each year. Such a large collection can produce duplicate functions, scattered logins, weak data practices, inaccessible design, and more work for teachers and families.
Schools need a clear reason for every tool. A product should solve a defined learning problem, fit teacher workflows, and show value beyond a polished sales pitch.
Five Checks Can Guide School Decisions
Mote proposed five quality indicators for educational technology. A tool should be:
- Safe: It protects student data and limits harmful interactions.
- Interoperable: It works with current school systems.
- Usable: Teachers, students, and families can use it without needless complexity.
- Inclusive: It supports accessibility and varied learning needs.
- Evidence-based: Research or measured outcomes support its use.
Her testimony recommended a Tier 4 evidence threshold within the federal education evidence framework as a minimum entry point for classrooms. She said fewer than half of purpose-built edtech tools meet that standard, compared with about 2% of consumer tools. Those comparisons also require checks against the underlying data and the exact federal framework.
Districts can test tools through limited pilots, review results each year, and remove products that fail to improve learning or safety. A pilot should measure student progress, teacher time, accessibility, privacy, and actual use.
Why Screen Value Matters More Than Screen Time
Screen Minutes Say Little About Learning
A student may spend ten minutes writing, practicing, or receiving accessibility support. Another may spend ten minutes passively watching content. The clock treats both activities as equal.
Mote urged schools to focus on "return on instruction": the learning benefit produced by a technology task. Leaders should ask whether students are thinking, creating, practicing, receiving useful feedback, or gaining independence. They should also ask whether an offline activity would work better.
Match the Tool to the Learning Task
Mote cited a claim that screen-based reading is six to eight times less effective for comprehension than reading physical books. That figure should be tied to its original study before publication or policy use. The comparison still supports a broader point: different tasks need different formats.
Reading long text may work best on paper for some students. Repeated decoding practice may work well with a focused digital tutor. Mote said decoding can require roughly 1,000 practice attempts for fluency and argued that purpose-built AI may help provide that repetition. Schools should judge the tool by the skill being taught, not by screen use alone.
Protect Students Who Need Digital Access
Blanket screen bans can harm students with disabilities who rely on digital tools for required accommodations. Assistive technology, speech tools, digital texts, and instructional programs should not be treated like recreational entertainment.
School policies should exempt approved accessibility tools and account for each student's learning plan. Special education staff and families should help shape these rules. The goal is better learning and access, not identical limits for every student.
How Risk-Based AI Policy Can Protect Students
High-Risk Systems Need Strong Controls
Mote proposed a "waterfall" approach to AI safety. The strongest controls would apply to high-risk systems, including public consumer platforms and chatbots designed to create emotional relationships.
These systems may raise concerns about manipulation, dependency, privacy, false information, harmful content, and non-consensual intimate imagery. Age checks, safety testing, monitoring, reporting channels, and enforcement should match those risks.
Purpose-Driven Tools Need Proportionate Oversight
An AI decoding tutor and an open-ended social chatbot do not pose the same risks. Educational tools still need privacy protection, age-appropriate design, accessibility, transparency, and evidence of learning value. Their oversight can focus on those specific needs instead of applying every rule meant for a public social platform.
Mote called for a fully staffed Office of Educational Technology, joint AI research through the NSF, IES, and NIH, protection for E-Rate funding, and support for the FTC and other agencies that address digital abuse. Federal research should examine learning outcomes, teacher training, student safety, privacy, access, and cognitive development.
Designing Schools Where AI Supports Human Judgment
Build AI Literacy Into Daily Instruction
Students need clear rules for checking claims, protecting personal information, sharing AI-assisted work, and recognizing hallucinations. Assignments can require source checks, original reasoning, reflection, and records of the work process.
Teachers also need time and training. Professional development should cover lesson design, bias, privacy, accessibility, classroom use, and false AI output. District policy should be shaped with teachers, students, families, special education staff, and technology teams.
Measure Learning, Safety, and Equity Together
Adoption rates do not prove success. Schools should track student learning, teacher workload, safety incidents, privacy concerns, accessibility, and unequal access.
A tool that saves time but weakens reasoning is not a strong investment. A tool that improves decoding for a student with limited access to one-on-one practice may offer high instructional value. These outcomes should guide future purchases and policy changes.
Conclusion
Generative AI is already in schools, but rapid adoption has outpaced training and risk review. Erin Mote's testimony calls for a better standard: choose tools that are safe, usable, inclusive, interoperable, and supported by evidence.
That means separating consumer chatbots from purpose-built learning systems, replacing screen time with screen value, and applying stronger controls to high-risk tools. It also means protecting accessibility, funding teacher training, and building federal research and enforcement capacity.
Schools should not have to choose between innovation and student safety. They should demand AI that strengthens learning while keeping human judgment, ethics, and student well-being at the center.
Post a Comment