Utah's SB322: A Practical Sandbox for Classroom AI - Without Giving Up Human Judgment
- SB322 creates an "Education Technology Regulatory Sandbox" to test AI in real classrooms before any statewide rollout.
- Participation is voluntary for districts, teachers, parents, and students, with clear opt-in/opt-out controls.
- Human-in-the-loop is mandatory: AI cannot assign final grades, make placements, or override teacher judgment without human review and approval.
- Student protections are explicit: no simulated romantic/personal relationships, clear disclosure when AI is used, due process, and no penalties for declining AI tools.
Utah is not banning AI in schools. It's putting it on a leash first.
SB322, approved by the Senate and awaiting a House vote, would stand up a time-limited, evidence-driven sandbox where schools can pilot AI under real classroom conditions. The goal is simple: allow innovation that helps teachers and students while keeping guardrails tight enough to protect kids and preserve professional judgment.
Why a sandbox now
AI is already in classrooms. Teachers are testing tools to save time; students are experimenting on their own; vendors are selling aggressively. Without a framework, districts risk adopting tools that expose students to inappropriate content, collect unnecessary data, or simulate emotional relationships with minors - problems that can trigger reactionary bans and chill useful innovation.
The bill aims to avoid that whiplash. It sets a controlled proving ground so the state can evaluate what works, what doesn't, and what should never be allowed anywhere near a student.
What SB322 actually does
- Voluntary participation for schools, educators, parents, and students, with opt-in/opt-out at every layer.
- Pre-pilot red teaming by vendors to surface failure modes, prompt injections, and boundary bypass attempts before tools touch classrooms.
- Human-in-the-loop by default: no AI-assigned final grades, placements, or decisions that affect a student's record without educator review and approval.
- Student dignity and due process: clear disclosure when AI is used; students can request human review of AI outputs; no penalties for declining AI.
- Hard lines on inappropriate behavior: AI systems may not simulate romantic or personal relationships with students.
- Evidence before expansion: independent evaluators run pre- and post-tests to verify outcomes before any statewide adoption.
- Legislative oversight for any statewide authorization (no administrative end-runs).
- Safe harbor for responsible research with privacy protections.
For educators: what this means in the classroom
If your school opts in, you remain the decision-maker. AI can draft, suggest, or flag - but you approve. Students know when AI is in play, can opt out, and can ask for a human check on anything that touches their record.
Used wisely, this setup can support reading and writing instruction, multilingual learners, and teacher workflow. Used carelessly, it can erode trust. The sandbox is built to test the former and block the latter.
For administrators: how to prepare now
- Set governance: define who approves pilots, evaluates risk, and sunsets tools that miss the mark. Align with the NIST AI Risk Management Framework for a common language on risk, controls, and measurement.
- Write clear opt-in/opt-out flows for families and staff. Keep records auditable. Make "human review" pathways obvious and fast.
- Procurement rules: require vendor red-team results, data minimization, and student-data isolation. No shadow profiles. No selling or training on student data.
- Instructional guardrails: define where AI can support (lesson planning, feedback drafts, reading supports) and where it cannot (final grades, placements, disciplinary decisions).
- Evaluation plan: pick a small set of measurable outcomes (e.g., reading fluency gains, teacher time saved, translation accuracy). Run pre/post. Share results.
- Professional learning: brief staff on safe prompts, bias checks, and human-in-the-loop habits. See AI for Education for practical classroom guidance and examples.
- School-level leadership: if you oversee pilots, map policy, workflows, and review gates. The AI Learning Path for School Principals can help build that operating system.
Addressing the big concerns
Studies have flagged risks to independent thinking and social-emotional development when AI is used poorly. A recent Common Sense Media report on AI companions found many teens experiment with chatbots for social interactions. SB322 draws hard boundaries against simulated relationships, builds transparency, and puts educators in control of consequential decisions.
On the opportunity side, targeted AI supports can help multilingual learners and reduce teacher busywork. The sandbox is the test bed to separate signal from noise - with proof required before anything scales.
What to watch next
The Senate has approved SB322. A final vote in the House will determine whether Utah launches the sandbox statewide as a structured pilot.
For now, the practical move is to get your house in order: governance, consent, evaluation, and training. If the bill passes, you'll be ready to participate on your terms - with your educators and your students protected.
Bottom line: AI can help, but only if humans stay in charge. SB322 is Utah's attempt to make that the default.
Your membership also unlocks:
