K-12Issue 144 records
K–12
Fresh K-12 signal pairs teacher-AI implementation evidence with homework-learning caution and hardening state controls. A UW/Colleague AI midyear report across 12 Washington districts finds teacher GenAI use exploding into lesson planning (60–99% of usage) with a preliminary +0.025 SD reading interim association (math null) among 12,005 students—explicitly correlational under voluntary adoption. Against that staff-copilot optimism, an ALEKS/UCI–McGraw Hill preprint shows post-ChatGPT AI-susceptible math study time collapsing (high school −31.3% cumulative) and ~25% lower odds on proctored retention items—U.S. platform evidence that faster homework completion can mask learning loss. On the control side, Florida’s Sept 18 FAR Change notice to Rule 6A-1.0957 requires agentic-AI kill-switches, activity logs, and bans AI as decision-maker without human review (grading/discipline/ESE referrals, etc.) ahead of a July 1, 2027 local policy deadline. VCU MERC’s Virginia comparative review shows why that floor matters: staff AI use was the thinnest domain—only two of ten divisions had explicit staff provisions—as SB394/HB1186 forces policy buildout. Together: evaluate teacher copilots on interim joins; evaluate student homework AI on unaided retention; ship staff allowlists and kill-switches before scaling student agents.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 2 academic research1 standards or public-body guidance1 independent research
- Outcomes
- 2 cautionary1 mixed1 emerging
- Source freshness
- 2 older, newly relevant1 recent1 new this fortnight
- Research completed
- 2026-09-20
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Teacher-facing AI shows modest correlational learning signals while homework-adjacent GenAI shows durable learning costs
Washington multi-district teacher AI use concentrated in lesson planning and associated with a small +0.025 SD reading interim correlation, while ALEKS platform evidence shows post-ChatGPT collapses in AI-susceptible study time and ~25% lower odds on proctored retention—implying staff copilots and student homework AI need different evaluation bars.
Operating questionAre we evaluating teacher planning copilots on interim learning joins while separately stress-testing student homework AI on unaided/proctored retention?
Supporting evidenceUniversity of Washington AmplifyLearn.AI and Colleague AI (IES/NSF-supported)UC Irvine and McGraw Hill (ALEKS platform researchers)
Binding state AI floors are arriving while local staff-use and procurement policies remain the thinnest domains
Florida’s Sept 18 Change notice hardens human-review, agentic kill-switch, and July 2027 policy deadlines, yet VCU MERC finds staff AI use was the least governed domain in studied Virginia divisions (only two of ten with explicit staff provisions)—so compliance work must prioritize staff allowlists and vendor training-data bans, not only student AUPs.
Operating questionDo our staff-use allowlists, kill-switches, and no-training-on-district-data clauses exist before student AI features scale under new state rules?
Supporting evidenceFlorida State Board of Education / Florida Department of Education (Rule 6A-1.0957)Metropolitan Educational Research Consortium (MERC), Virginia Commonwealth University School of Education
Evidence ledger
Full record · every source keeps its link and limitations
UW/Colleague AI midyear report: WA teacher AI use correlated with +0.025 SD reading gains; lesson planning dominates
Across 12 Washington districts, teacher use of Colleague AI grew from 1,438 teachers in September 2025 to 2,891 by end-December (412,304 messages), reaching 4,644 teachers by March 31, 2026 (+61%). Active users averaged ~12 sessions and 3.7 hours; Claire AI lesson-planning features comprised 60–99% of usage in every district. Teachers serving more ELL/SPED students referenced differentiation and accessibility features more often. Among 12,005 students with matched BOY–MOY scores, roughly three hours of teacher AI use associated with a statistically significant +0.025 SD reading interim gain; math association was not distinguishable from zero. Authors label the student-outcome link a preliminary correlation under voluntary adoption.
Why it matters, evidence and limitations
- Why it matters
- Shows where teacher-facing GenAI actually concentrates (lesson planning) and how modest correlational ELA signals—not math—appear in multi-district U.S. K-12 settings, informing SEA/LEA expectations for staff AI ROI and evaluation design.
- Evidence and measured results
- Opened arXiv PDF end-to-end (executive summary key findings, adoption tables, feature-mix figures, BOY–MOY correlation section). Cross-checked HTML mirror for continuous wording. Confirmed authors’ caution that only a subset of districts provided mid-year admin data and that findings linking use to student characteristics are preliminary signals.
- Limitations and uncertainty
- Correlational, voluntary adoption, first four months; vendor coauthor (Colleague AI); subset of districts supplied admin data; not an RCT; effect size small; math null. Why now: first unarchived multi-district WA teacher-AI↔student-score package contemporaneous with binding state AI rules and homework-learning-penalty evidence.
ALEKS panel: post-ChatGPT AI-susceptible math study time falls sharply; proctored retention odds drop ~25%
Using a decade of ALEKS logs (3.2M learning interactions; 12.2M PPL placement responses), authors exploit AI-susceptible text word problems vs AI-resistant graph tasks. After ChatGPT’s release, learning time on AI-susceptible problems fell ~2.8% per quarter for college students (26.9% cumulative over 11 quarters); high school showed 31.3% cumulative decline, middle school 9.0%, Grade 5 null. Among college students the post-ChatGPT time divergence vanishes under proctoring. Logistic models on randomly assigned proctored retention items show ~25% cumulative decline in odds of correct response, while the same estimator on non-proctored assessments shows a large opposite-signed increase—evidence authors interpret as AI-altered study behavior with durable learning costs.
Why it matters, evidence and limitations
- Why it matters
- Provides U.S. platform-scale behavioral evidence that homework-adjacent GenAI can shrink time-on-task and unaided retention—directly informing K-12 assessment integrity, AI homework policies, and parent/PTA debates after recent SEA/LEA AI fights.
- Evidence and measured results
- Opened arXiv PDF abstract, significance statement, methods (topic susceptibility design, PPL proctoring contrast), and main cumulative effect estimates for high school/middle school/Grade 5. Confirmed preprint-under-review status and McGraw Hill coauthorship.
- Limitations and uncertainty
- Preprint under review; McGraw Hill coauthors on ALEKS vendor platform; college PPL proctoring contrast is cleaner than K-12 classroom settings; Grade 5 null and smaller middle-school effects limit elementary claims. Why now: newly frames U.S. homework-AI risk alongside prior China secondary outsourcing evidence and Sept 2026 district AI policy fights—not previously in the k12 archive.
Florida FAR Change notice (Sept 18): AI never decision-maker without human review; agentic tools need kill-switch
On September 18, 2026, Florida published a Notice of Change to Rule 6A-1.0957 clarifying post-hearing amendments: definitions expanded to generative and autonomous/agentic AI instructional tools; districts/charters must amend internet-safety policies by July 1, 2027. The change requires that agentic/autonomous AI in VPK–12 be used only under direct adult supervision, maintain complete activity logs, and include a mechanism for school personnel to immediately suspend/disable autonomous functionality without vendor assistance. It also requires that AI never be used as a decision maker absent human review for grading, evaluations, promotions, discipline, threat assessment, counseling/ESE referrals, or communications, and that districts disclose unexpected autonomous behavior or IT safety/security incidents from approved tools to the Department. Broader proposed package (opened companion notice) still includes parent notification/opt-in with substantively comparable non-AI alternatives, PreK–5 developmental review, bans on companion/anthropomorphic relationship design, and staff training.
Why it matters, evidence and limitations
- Why it matters
- A binding statewide operating floor for student AI tools—parent voice, companion bans, human-in-the-loop decision bans, and vendor kill-switches—that other SEAs/LEAs will copy or contrast with guidance-only models.
- Evidence and measured results
- Opened official FAR Change notice DOC (ID 31384852) via flrules notice_Files and extracted full amendment text; cross-checked View_notice HTML (publish date 9/18/2026) and the Aug 26 proposed-rule notice DOC (ID 31301432) for parent opt-in/companion/training provisions; noted same-day WCTV coverage but cited the official notice as verification object.
- Limitations and uncertainty
- Change notice amends proposed text; final FAC codification timing may lag; higher-ed portions of political messaging are out of k12 stream scope here—we cite K-12 instructional-tool provisions only. Tallahassee Democrat Sept 16 coverage already archived—this edition uses the Sept 18 official Change notice as the fresh primary.
VCU MERC: staff AI use is thinnest VA division domain—only 2 of 10 had explicit staff provisions
MERC analyzed publicly available AI-related policies across Virginia divisions against six domains (student use, academic integrity, staff use, approved tools/procurement, data privacy/security, training/literacy) as Virginia’s SB394/HB1186 (signed April 2026; effective July 1, 2026) requires VDOE guidance and eventual local board policies. At data collection, six of ten studied divisions had no AI-explicit language. Staff Use of AI had the least existing governance infrastructure—only two divisions had explicit staff AI use provisions—with no reliable Acceptable Use Policy equivalent for teacher AI. Three divisions had AI procurement/vetting language; only one tied procurement to privacy agreements and a ban on using division data for model training. Every division had some privacy framework, but FERPA-era policies were not designed for AI model-training risks. Training/literacy was stronger at state level than in local requirements. Why now: documents the staff-use/procurement gap LEAs must close as binding state AI mandates (VA and FL) land—unarchived and newly relevant to Sept 2026 compliance work.
Why it matters, evidence and limitations
- Why it matters
- Gives SEAs/LEAs a domain checklist and empirical gap map—especially staff-use and procurement—usable when translating new state AI laws into board policy and vendor contracts.
- Evidence and measured results
- Opened full VCU Scholars Compass PDF (recommended citation, executive findings on six-of-ten and staff-use domain, procurement and privacy bullets). Cross-checked July 8, 2026 VCU News interview summarizing the same gaps and the July 1 law’s compliance shift.
- Limitations and uncertainty
- Ten-division comparative sample, not statewide census; publicly posted documents only—may miss internal drafts; June/spring 2026 snapshot precedes full VDOE guidance issuance. Companion GMU 50-division preliminary (38% with formal policies) not used as primary to avoid double-counting Virginia gap narratives.
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.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.
- Independent research
- Research conducted outside the implementing organization.