K-12Issue 114 records
K–12
Fresh K-12 signal shifts from yesterday’s tutoring RCTs to adoption mechanics, instructional quality, and privacy verification. A September 2026 national math/science teacher survey finds GenAI use rising (58%, 31% frequent) without shorter workweeks—frequent users spend more planning time—while training and district guidelines still lag. Instruction Partners’ AI in Action Learning Tour (fieldwork across 16 systems; findings pages updated 2026-09-16) warns that general-purpose chatbot access during assignments can significantly damage effortful learning, even as purpose-built tools show conditional promise under teacher leadership. AP/EdWeek coverage of the Microsoft-AFT National AI Safety & Privacy Standard (effective for Microsoft-contracted schools Nov 1) raises a contractual privacy floor that Google has not matched. Class Size Matters’ Sept 15 NYC Council testimony supports the city’s student AI pause but argues ERMA vetting remains checklist-weak with exception and pilot loopholes. Together the sources urge districts to couple contracts and moratoria with instructional criteria, PD, inventories, and independent verification—not license counts or press releases alone.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 2 independent research2 independent reporting
- Outcomes
- 2 mixed1 emerging1 cautionary
- Source freshness
- 3 new this fortnight1 recent
- Research completed
- 2026-09-17
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Contractual AI privacy floors matter only if districts can inventory and independently verify tools
Microsoft-AFT reporting advances enforceable no-training/audit privacy terms, while NYC testimony argues large-district vetting remains checklist-based with weak post-breach investigation. Together they show SLED buyers need both stronger vendor contracts and operational verification capacity—especially across non-signatory vendors and exception cohorts.
Operating questionHow will we attach Microsoft-class privacy terms and independently verify AI features across our full edtech portfolio, including exception pathways?
Supporting evidenceEducation Week (Associated Press)Class Size Matters / Parent Coalition for Student Privacy
GenAI adoption is rising in unstructured ways that threaten effortful learning
The national teacher survey shows rising teacher GenAI use without workload relief, widespread missing training/guidelines, and mixed views of student impact—especially academic integrity. Instruction Partners’ tour concludes general chatbot access during assignments can significantly damage learning and that only carefully designed, teacher-led uses add value. Districts should pair access decisions with instructional design and PD, not license growth alone.
Operating questionWhat access controls, instructional criteria, and PD will we deploy so GenAI increases effortful learning rather than substituting for it?
Supporting evidenceAmplifyGAIN Center / University of Washington (with WestEd and Colleague AI)Instruction Partners
Evidence ledger
Full record · every source keeps its link and limitations
National math/science teacher survey finds GenAI adoption rising without workload relief or clear policy
In a 2026 national survey of U.S. K-12 math and science teachers (N≈1,011), GenAI use for instructional tasks rose to 58% (31% weekly/daily) from 50%/21% in 2025. Use remains mostly teacher-facing—76% of users cite instructional planning—while 74% do not include student GenAI in class. Frequent users report about 45-hour workweeks similar to non-users and spend 7.3 hours/week on lesson planning, 1.4 hours more than non-users. 58% report no formal GenAI training and 42% report no district GenAI guidelines, though formal guidance under development rose to 31% from 18%. Teachers rate student-learning impacts mixed (29% positive, 32% negative) with strongest negativity on academic integrity (60%).
Why it matters, evidence and limitations
- Why it matters
- Directly informs U.S. district and state K-12 buyers evaluating whether GenAI licenses reduce teacher workload or improve learning. Shows adoption is growing faster than training, policy clarity, and measured student benefit—relevant for procurement, PD design, and operating-model choices that distinguish assisted teacher output from retained student learning.
- Evidence and measured results
- Self-reported survey of public-school math/science teachers fielded via RAND American Teacher Panel–style methods described in the report; 2026 N=1011 unweighted counts in charts; compares 2025 (N=979) to 2026. Associations between usage intensity and time use are correlational, not experimental assignment.
- Limitations and uncertainty
- Math/science teachers only; self-report bias; usage groups not randomly assigned; no direct student learning outcome measures; Colleague AI co-authorship creates potential interest in teacher AI tools.
Instruction Partners tour warns general chatbots damage effortful learning while purpose-built tools show conditional promise
After Year 1 fieldwork—32 classroom observations in 16 systems, 50+ teacher and 100+ student interviews across 20 student-facing AI products—Instruction Partners concludes teachers remain pedagogical authorities; well-designed student-facing AI can add practice and feedback; many AI uses underperform paper-pencil instruction and feel isolating/boring; if students can use general-purpose chatbots while completing assignments, learning will be significantly damaged by reduced effortful thinking; strongest experiences come from products with a clear division of teacher vs AI roles rather than maximally flexible tools.
Why it matters, evidence and limitations
- Why it matters
- Gives SLED districts an instructional quality lens beyond privacy checklists: product selection should favor purpose-built designs with effortful-thinking safeguards, limit general chatbots during assignments, and keep teachers central—aligning with developmental-risk and retained-learning concerns in the K-12 brief.
- Evidence and measured results
- Qualitative field study with classroom observations and structured user interviews; non-random product sample chosen from partner use and landscape scan; authors disclose they did not evaluate privacy, SIS integration, pricing, or environmental impact; not a causal RCT.
- Limitations and uncertainty
- Non-comprehensive product sample; early-adopter sites may differ from typical districts; instructional lens only; white paper download gated; mixed qualitative evidence not causal impact estimates.
AP/EdWeek: Microsoft-AFT school AI privacy pact sets contractual bar; Google has not signed
Reporting dated September 15, 2026 covers Microsoft’s legally binding National AI Safety & Privacy Standard negotiated with AFT/UFT: no student/educator data for AI training (narrow safety exception), no sale/ads/product-development use, bans AI companions designed to foster emotional dependency, requires third-party audits and plain-language family disclosures, applicable to Microsoft-contracted schools starting November 1. OpenAI and Anthropic say they are discussing similar pacts; Google has not said it will offer comparable protections. Experts caution the framework could be misread as a green light to adopt tools and highlight inequities when districts lack bargaining power—while NYC and LA pause much student AI use.
Why it matters, evidence and limitations
- Why it matters
- Gives SLED procurement teams a concrete contractual template and competitive pressure point for vendors, while warning that privacy standards do not answer age-appropriateness or instructional-value questions.
- Evidence and measured results
- AP reporting published via Education Week quoting Microsoft, AFT, Fairplay, and Center for Democracy & Technology; describes agreement terms and marketplace response. Not an independent legal audit of contract text.
- Limitations and uncertainty
- Journalistic summary; full legal enforceability depends on district contract language; Google non-participation limits systemwide coverage; does not measure learning outcomes.
Class Size Matters tells NYC Council ERMA AI vetting remains checklist-weak despite student AI pause
In September 15, 2026 NYC Council testimony, Class Size Matters praises the city’s school-year pause on most student-facing generative AI (DOE claims ~40 products disabled) but argues oversight remains inadequate: DOE historically lacked inventory of which edtech uses AI; ERMA vendor vetting is described as a self-filled checklist with little independent verification; breach response has relied on vendors (Illuminate example). Testimony flags loopholes assigning AI to students with disabilities/ELLs when deemed “necessary,” high-school AI literacy/CTE mandates, and pilots of Quill/Edia/Brisk/Playlab for up to ~50,000 students without clear independent evaluation or parent opt-out; teacher use remains largely unrestricted beyond unclear grading/high-stakes limits. Calls for independent privacy, security, bias, and peer-reviewed educational-benefit audits.
Why it matters, evidence and limitations
- Why it matters
- Provides contradictory scrutiny to vendor privacy marketing and large-district AI pauses: operational inventory, independent verification, and exception governance determine whether moratoria actually reduce developmental and privacy risk.
- Evidence and measured results
- Advocacy testimony citing NYC DOE guidance, NY State Comptroller student-data audit, and prior Illuminate breach reporting. Not a randomized study; positions reflect Parent Coalition/AI Moratorium Coalition agenda—useful as scrutiny, not as sole fact base.
- Limitations and uncertainty
- Advocacy source with campaign stance; some DOE responses may be incomplete rather than final; quantitative claims about pilot scale should be verified against DOE primary documents.
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 research
- Research conducted outside the implementing organization.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.