K-12Issue 385 records
K–12
Tonight's K-12 signal is a sharper line between AI that helps students perform and AI that helps them learn, plus the governance plumbing to enforce it. Australia's national evidence body, AERO, released paired explainers today. They warn of a 'performance paradox', where students do better on a task with AI but learn less, along with cognitive offloading and 'illusions of knowledge'. They advise extra caution before secondary school and tell policymakers to fund equitable access to quality tools and real evaluation, because evidence is short-term and mixed once AI is removed. Oklahoma's updated model board policy turns governance into concrete obligations: approved tools only, an AI inventory, annual parent notices, opt-out with non-AI alternatives, and a public AI Impact Report, with full compliance due in 2027–28. England's Confederation of School Trusts offers boards a matching assurance method: survey staff for shadow AI, keep a use-case register, and test supplier claims against evidence. Common Sense Media's October 13 response to OpenAI adds that ChatGPT for Teens parental alerts can take up to three hours to activate, a delay disclosed only after testing, and failed even past that window. That is a reminder that vendor safety features need independent verification.
- Evidence records
- 5
- Cross-source patterns
- 3
- Evidence classes
- 3 standards or public-body guidance1 independent research1 public-sector association guidance
- Outcomes
- 2 cautionary2 emerging1 mixed
- Source freshness
- 5 new this fortnight
- Research completed
- 2026-10-15
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Evidence bodies and boards converge on a learning-first test for approving AI, not task performance or supplier claims
AERO says AI can raise task performance while reducing learning and asks schools to judge tools by whether they support the thinking learning requires; its policymaker brief adds that free tools trail purpose-built ones. CST tells boards to test supplier claims against research evidence. Together they push districts to require evidence of unassisted learning before approval.
Operating questionDoes our AI approval process ask for evidence of learning without AI, or only for engagement, usage or task scores?
Supporting evidenceAERO explainer for school leaders and teachers: AI can create a ‘performance paradox’; prioritise evidence-based teaching, guardrails and extra caution before secondary schoolAERO explainer for policymakers: base AI policy on learning science, fund equitable access to quality tools, and invest in monitoring and evaluationConfederation of School Trusts (CST)
An AI inventory or register is becoming the baseline governance artefact on both sides of the Atlantic
Oklahoma's model policy requires a current AI tool inventory, annual parent notices of tools, vendors and data, and a public AI Impact Report. CST asks trust boards for an AI use-case and systems register with owners, data, oversight and review dates, plus a shadow-AI survey. Both treat visibility as the precondition for every other control.
Operating questionCan we list every AI tool and embedded AI feature in use, with owner, data involved and review date, today?
Supporting evidenceOklahoma State Department of Education (OSDE)Confederation of School Trusts (CST)
Vendor safeguards for students need independent verification and design guardrails, not marketing assurances
CSM reports ChatGPT for Teens alerts failed even after the activation delay OpenAI disclosed only after testing. AERO says AI's effect depends on whether tools build in pedagogical guardrails and urges age restrictions and teacher oversight for younger students. Districts should test safety and learning guardrails themselves before relying on them.
Operating questionBefore telling families a student AI tool is safe, have we tested its crisis and learning guardrails on newly provisioned accounts?
Supporting evidenceCommon Sense Media Youth AI Safety InstituteAERO explainer for school leaders and teachers: AI can create a ‘performance paradox’; prioritise evidence-based teaching, guardrails and extra caution before secondary school
Evidence ledger
Full record · every source keeps its link and limitations
AERO explainer for school leaders and teachers: AI can create a ‘performance paradox’; prioritise evidence-based teaching, guardrails and extra caution before secondary school
AERO's 12-page explainer (October 2026) gives provisional guidance to Australian school leaders and teachers. It states there is limited research on AI and student learning outcomes in schools, mostly from secondary and higher education. It warns that student AI use can create a ‘performance paradox’ (better task performance, less learning), cognitive offloading, over-reliance that weakens metacognition, and ‘illusions of knowledge’. It sets five considerations: prioritise evidence-based teaching, monitor risks to learning, set clear expectations for AI use, rethink assessment and feedback, and apply a culturally responsive lens. It advises greater caution, including age restrictions and teacher oversight, before secondary school, and says teachers keep responsibility for assessment judgements. Why now: released today and the first national evidence-body guidance in this archive that makes the assisted-performance versus retained-learning distinction the organising principle.
Why it matters, evidence and limitations
- Why it matters
- Directly addresses the K-12 brief's assisted-output versus retained-learning distinction. U.S. districts and state agencies drafting guidance can borrow its structure (learning-science first, guardrails, primary-grade caution, assessment validity) while noting it is Australian and provisional.
- Evidence and measured results
- Evidence synthesis and guidance, not a new trial. Cites Bastani et al. (PNAS) for unassisted-exam harm, Fesler et al. (2026) K-12 evidence review, Yan et al. (2025) and others. Lists limitations: most studies recent, short term, exploratory and outside Australia; many measure engagement or perceptions; very limited primary-school research; little evidence on durability once AI is removed.
- Limitations and uncertainty
- Provisional guidance from an Australian evidence body; relies on an evidence base it describes as limited and mostly non-Australian. Not an evaluation of any product. U.S. legal frameworks (FERPA/COPPA, state laws) are not addressed. The PDF sits behind a JavaScript client challenge for plain HTTP clients; content was retrieved with a page-rendering fetch tool.
AERO explainer for policymakers: base AI policy on learning science, fund equitable access to quality tools, and invest in monitoring and evaluation
AERO's 14-page companion explainer for policymakers (October 2026) states evidence on AI impact in schools is limited and that results are mixed when students are later assessed without AI, citing the Bastani et al. high-school maths study where unrestricted AI users scored lower on an unassisted exam. It lists opportunities (teacher time on planning and admin, tailored resources, tutoring support) and risks (cognitive offloading, assessment validity, privacy and data governance, inequity). It notes free general-purpose tools have lower capabilities than subscription models and are less aligned to curriculum, which may widen gaps in rural, remote or disadvantaged communities. It recommends three actions: clear evidence-informed policy and guidance; implementation support through access to quality tools and workforce capability; and sustained research, monitoring and evaluation. Why now: released today alongside the school-leader explainer.
Why it matters, evidence and limitations
- Why it matters
- Gives state education agencies a system-level template: equitable access to quality tools, workforce capability, and funded evaluation, rather than leaving each district to choose consumer tools. Relevant to U.S. states building approved-tool programs and guidance updates.
- Evidence and measured results
- Evidence synthesis, not original research. Cites Bastani et al., Fesler et al. (2026), Pearson & Jensen (2026), Roy et al. (2024) on lesson planning, and Miao & Holmes (2023). Does not quantify effects itself; describes the evidence base as limited and evolving.
- Limitations and uncertainty
- Australian system context; recommendations are general and not costed. No new outcome data. Retrieved through a page-rendering fetch because the PDF is behind a JavaScript client challenge for plain HTTP clients.
Oklahoma SDE updated model board AI policy (Oct 2026): approved-tool review, AI inventory, annual parent notification, opt-out rights and full compliance by 2027–28
OSDE posted an updated model board policy, 'Artificial Intelligence (AI) Use in Schools', in October 2026 for districts implementing the 2026 Oklahoma Responsible Technology in Schools Act (70 O.S. § 11-120). The four-page template defines classroom, student-facing, administrative and third-party vendor AI and requires human-in-the-loop review. It assigns roles (board, superintendent, technology department approval of tools, educators, students) and lists approved and prohibited educator uses; prohibited uses include AI as the primary basis for grades, discipline or placement, deepfakes, and entering sensitive student data into consumer-grade tools. It requires data minimisation, district-approved tools only, a current AI tool inventory, vendor FERPA/COPPA compliance, annual parent notification of tools, vendors, data and purpose, 30-day notice for new data-collecting tools, opt-out from any student-facing AI tool with non-AI alternatives, annual review, a public AI Impact Report, and full compliance by the start of 2027–28. Why now: the October update follows Oklahoma Watch's archived Oct 5 report on uneven district implementation and gives the primary template districts are adopting.
Why it matters, evidence and limitations
- Why it matters
- A concrete U.S. state template that turns AI governance into board-adoptable obligations with a compliance date. Gives vendors and districts a checklist (inventory, notification, opt-out, impact report) likely to recur in other states.
- Evidence and measured results
- Primary policy template, not an evaluation. Requirements are stated obligations; there is no data on adoption or effectiveness. Bracketed fields indicate districts customise it.
- Limitations and uncertainty
- Model template only; districts may alter it. Not evidence that policies are implemented or effective. The liability clause's enforceability is not established. Retrieved PDF metadata shows October 1 creation; the agency page labels it updated October 2026.
Common Sense Media rebuts OpenAI on ChatGPT for Teens: alerts failed even on accounts linked over three hours; OpenAI disclosed the delay only after testing
On October 13, CSM's Head of AI & Digital Assessments responded to OpenAI's claim that the bulk of testing for the October 7 'Unacceptable Risk' rating may have preceded activation of parental controls. CSM says it confirmed with OpenAI before testing that Teen mode features were launched on its accounts, and that almost all testing ran on accounts in use for weeks. In a separate test, new accounts were linked to parent accounts before the first message and testers disclosed suicidal thoughts, self-harm or disordered eating; none produced a notification. After testing, OpenAI disclosed alerts can take up to three hours to switch on after linking; CSM says some alert tests ran past that window and still produced no alerts, the interface showed accounts as linked immediately, and OpenAI updated its Help Center only afterwards. CSM lists retest criteria. Why now: new information after the archived assessment, including the vendor-disclosed activation delay and the vendor's dispute.
Why it matters, evidence and limitations
- Why it matters
- Districts that tell families consumer teen modes include parental alerts need to know the alerts may not be active on linking and were not reliable in independent tests. Relevant to student-safety communications, approved-tool decisions and counselling protocols.
- Evidence and measured results
- Tester's own account of its methods and logs; no published per-test log or counts in this post. The underlying assessment (archived separately) reports more than 4,000 prompts. The OpenAI position is described second-hand by CSM and media.
- Limitations and uncertainty
- Advocacy-aligned tester responding to a vendor dispute; claims are self-reported without raw logs in the post. OpenAI's full response is not included. Applies to a consumer product, not a district-licensed tool. Follow-up to an assessment already in the archive; this record covers only the new disclosures and dispute.
Confederation of School Trusts: AI is a board governance issue; keep an AI use-case register, find shadow AI, and test supplier claims against evidence
CST's 20-page guide (Sadeghi and Wathen, October 2026) argues AI is a governance issue, not just technology, because suppliers embed AI into products trusts already use. It says boards need stewardship, strategic oversight and assurance rather than technical expertise, and proposes an assurance framework within existing structures. Steps include defining AI scope (generative AI, agents, AI in existing platforms), finding 'shadow AI' through a low-stakes, no-judgement staff survey, and keeping a proportionate AI use-case and systems register (purpose, owner, users affected, data, supplier, human oversight, risks, assessments, approval status, review date) cross-referenced with the information asset register and Record of Processing Activity. Use cases are classified by impact, AI is integrated into existing policies, and supplier claims must be tested against research evidence, data security and pedagogical priorities. It references DfE generative AI product safety standards and filtering and monitoring standards, and includes procurement questions. Why now: published this month and not previously archived.
Why it matters, evidence and limitations
- Why it matters
- Board-level governance mechanics (register, shadow-AI discovery, proportional classification) map directly to U.S. school boards and charter management organisations that lack AI policies, as recent archived state surveys show.
- Evidence and measured results
- Sector guidance from a membership body representing trusts educating more than four million children; not an evaluation. No data on adoption or effectiveness of the framework.
- Limitations and uncertainty
- English legal and regulatory context (DfE standards, UK GDPR, KCSIE); U.S. transfer requires mapping to FERPA, COPPA and state laws. Guidance, not evidence of outcomes.
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.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.
- Independent research
- Research conducted outside the implementing organization.
- Public-sector association guidance
- Practitioner guidance or an association-supplied case; not independent outcome evidence.