K-12Issue 464 records
K–12
On the day of Maryland's AI policy deadline, Frederick County Public Schools adopted Regulation 124-01, turning its student AI ban into detailed rules. Student use requires an approved course, literacy lesson or IEP, tools must sit behind district sign-in with monitored inputs and outputs, staff may not grade with AI or rely on detectors, and vendors must meet ten procurement criteria and contract limits on data use and model training. Measurement remains the weak point: NPR reports Dayton and Wichita can only gauge AI pilots by perception, while Incheon, South Korea is building a comparison of policy and regular schools into its AI lesson model. In Kansas, a state grant funded human-verified AI gun detection for a 325-student district, according to the vendor. No source reports measured learning outcomes.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 2 independent reporting1 standards or public-body guidance1 vendor claim
- Outcomes
- 3 emerging1 cautionary
- Source freshness
- 4 new this fortnight
- Research completed
- 2026-10-23
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Human review is being written in before AI output can affect a student
Frederick County requires human oversight of any AI output in student or staff work and bars AI from making consequential decisions alone; ZeroEyes routes every detection through human reviewers before alerts reach police. Both put a person between AI output and action, though one is district rule and the other a vendor's service model.
Operating questionFor each AI system that can affect a student, who reviews the output before action, and is that review documented?
Supporting evidenceFrederick County Public Schools (Maryland)ZeroEyes, Inc. (vendor) and Elkhart USD 218
AI pilots are scaling ahead of outcome measures
Dayton's chatbot pilot could only measure perception and Wichita is holding pilots small until value can be measured; Incheon reports participation, perception and satisfaction figures while its comparison experiment is still planned. Evaluation design is lagging rollout.
Operating questionDoes each AI pilot have a defined outcome measure and comparison before launch, beyond satisfaction surveys?
Supporting evidenceNPR (All Things Considered)Incheon Metropolitan City Office of Education (reported by Electronic Times / etnews)
Evidence ledger
Full record · every source keeps its link and limitations
Frederick County Public Schools (MD) adopts Regulation 124-01 implementing its student AI ban: approval paths, no AI grading, no detector-led discipline, vendor contract terms
On October 22, 2026, the day of Maryland's AI Ready Schools Act deadline, FCPS adopted Regulation 124-01 and revised Policy 124 to reference it. The regulation keeps the PreK-12 student ban on AI products but defines approval paths: courses where the curriculum office says AI is required, AI literacy lessons, IEP or 504 plans when other assistive technology is unavailable, and other uses approved by curriculum and IT. Approved tools must be on the Approved Digital Tool List and reachable only after sign-in to the FCPS network, with inputs and outputs subject to monitoring. Prohibited uses include entering student or employee PII without a data-sharing agreement, AI making consequential decisions on its own, and staff grading learning tasks with AI outside approved computer-scored platforms. Staff may not rely on AI detectors as the primary integrity tool. Annual AI literacy is required for grades 2-12 and annual role-specific staff training. Procurement must evaluate ten criteria, including whether FCPS data trains vendor models and total cost of ownership; contracts must limit data use and prohibit advertising, profiling and unrelated model training. Operational AI affecting student services must be bias-evaluated before adoption.
Why it matters, evidence and limitations
- Why it matters
- The most detailed primary implementation text yet from a Maryland district at the state deadline; it turns a headline ban into specific access, procurement, oversight and incident-reporting rules that vendors and integrators will have to meet.
- Evidence and measured results
- Primary regulation text, opened in full. It is a rule set, not an evaluation; no data on usage, approvals or outcomes.
- Limitations and uncertainty
- Adopted text only; how approvals, monitoring and enforcement work in practice is not yet known. The regulation's history lists adoption on 10/22/26, but no board minutes or press coverage of the adoption were opened. The earlier Policy 124 adoption (October 7) is already archived; this record covers the new regulation, which is a separate document.
NPR: Dayton and Wichita AI experiments can only measure perception, and CRPE warns innovation is outrunning evidence-building
NPR reports on a chatbot named Jordan used last school year at the Alternative School at Jackson Center in Dayton, Ohio, which asks students for advice to draw out their feelings and grades their responses. The consultant who helped design and evaluate it says every attempt at concrete measurement only reached perception, and the district relies on asking students and staff. He cautions against scaling it as a full solution. Wichita Public Schools' AI specialist says the district is introducing AI tools slowly, including a high school AI math tutor, and that experiments will stay small until educators can measure their value. Robin Lake of the Center on Reinventing Public Education says innovation is proceeding without much attention to evidence-building and that states and the federal government should set guardrails rather than stop experimentation.
Why it matters, evidence and limitations
- Why it matters
- Direct evidence that districts lack outcome measures for AI pilots, including a student-facing social-emotional chatbot in a school for students with behavioral challenges, which limits scaling decisions.
- Evidence and measured results
- Broadcast report with interviews; no evaluation data, sample sizes or methods published.
- Limitations and uncertainty
- Published October 19 and not previously archived; anecdotal interviews; no quantitative results; the chatbot vendor and data handling are not described.
Incheon (South Korea) education office tests a 'think first, then AI, then verify' lesson model in 10 policy schools against comparison schools, with a planned grade 7 field experiment
The Incheon Office of Education said it is evaluating its 'Ilgeotsseu AI' (read-walk-write with AI) lesson model in 10 policy schools, from kindergarten through high school and special schools, against regular schools, with a Hanyang University team led by Professor Cho Byung-young. The model applies a human-AI-human sequence: students think first, use AI to widen questions and search, then verify and finish in their own words. So far 2,410 students have joined 23 projects averaging 9.6 lessons each, using tools including Padlet, Canva and Gemini. In a survey of 1,038 Incheon students, 62.7% of elementary and 71.3% of secondary students said AI is smarter than they are. In a pre-survey, 94.7% of 265 policy-school teachers said they use AI for school work, and 90.4% of training participants were satisfied. This semester the team plans a field experiment with about 120 grade 7 students comparing the model with general AI use, measuring AI literacy and critical use.
Why it matters, evidence and limitations
- Why it matters
- A system-level example of building a comparison design into an AI instructional rollout, relevant to U.S. districts that lack outcome measures; transferability is limited by Korea's centralized system and curriculum.
- Evidence and measured results
- Newspaper report of an education office announcement with participation and survey figures; no outcome results yet; experiment is planned, not completed.
- Limitations and uncertainty
- Secondary report in Korean; translation by the agent. Survey figures are self-reported; satisfaction is not an outcome; the planned experiment of about 120 students is small and results are expected only in 2027. The article also cites OECD and Turkish study findings secondhand; those are not used here.
Elkhart USD 218 (KS) deploys ZeroEyes AI gun detection with human-verified alerts, funded by a Kansas Attorney General grant (vendor release)
ZeroEyes announced that Elkhart USD 218, a rural Kansas district of about 325 students and 55 staff in three buildings, has deployed its AI gun detection software. The district is also installing a new camera system and upgraded card-reader access controls. It paid for the software with the Kansas Safe and Secure Firearm Detection Grant Program, administered by the Office of the Kansas Attorney General. The software runs on existing digital cameras; detections are sent to the vendor's operations center, where staff who are military and law enforcement veterans decide whether a threat is valid before alerting police and school security. The release quotes the district IT and athletic director on easy integration. It reports no detection, false-alert or response-time data for the district.
Why it matters, evidence and limitations
- Why it matters
- Shows a state grant program funding AI surveillance in very small districts, where one staff member may hold IT and athletics roles and depends on the vendor for monitoring.
- Evidence and measured results
- Vendor press release; company performance claims are not independently verified and no district data is reported.
- Limitations and uncertainty
- Vendor marketing; claims such as SAFETY Act designation and accuracy are the company's own. Independent reporting elsewhere has documented false alerts for AI weapons detection systems in schools, but none is reported for this deployment.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.