People · Practice · Possibility
Independent research for education leaders
Follow a question across the research. Search findings, sources, and the implications for your schools.
39 stories · Clear filters
The American Psychological Association distinguishes engagement and assisted performance from durable learning. It calls for evidence that knowledge transfers beyond an app or chatbot.
Adding mindful chat, breathing, and supportive messages produced no statistically significant advantage in algebra learning gains or math anxiety over cognitive tutoring alone.
Across 26 studies, AI supported reflection and feedback in some settings, while uncritical reliance was associated with reduced analytical engagement and weaker checking of information.
Only 20% of educators reported extensive AI training. Among parents, 77% wanted input into classroom AI use, while 20% clearly understood the guidance their child received.
State AI guidance emphasizes ethics and data privacy. Stakeholder trust and ongoing monitoring receive less developed treatment; equity and educator capacity receive moderate attention.
The UNICEF report estimates technology-facilitated sexual abuse affected about one in five internet-using 12-17-year-olds across the studied countries. More than four in ten cases were disclosed to no one.
Students produced ambitious AI-assisted work, but an unaided exam exposed weak performance. The course did not reliably demonstrate its foundational learning goals.
Students described varied AI uses, from feedback and explanations to extensive delegation. Interviews also revealed anxiety about unclear rules and false misconduct accusations.
Students described checking chatbot feedback, explaining their reasoning, and adapting learning strategies. They also reported unclear feedback and usability problems.
All four teacher-created app proposals specified meaningful evidence of student thinking. Only two explicitly described teacher authority, and only two named a safeguard.
Fairfax County Public Schools used AI to create a Python script that organized more than 40,000 employee records. Indianapolis Public Schools reports saving over $300,000 by building internal AI tools and expects another $500,000 in savings next year.
ChatGPT for Teachers expanded to 55 additional school systems, reaching over 300,000 educators across more than 100 K–12 organizations. OpenAI also introduced a Student Data Privacy Consortium agreement covering 15 states, including Virginia, plus California.…
For eligible K–12 students whose districts already enabled Copilot Chat, Microsoft now preselects its Study and Learn agent. Administrative enabling and age controls still apply.
Santa Fe approved a cybersecurity policy but deferred its student-data policy. The deferred draft proposed prohibiting vendor model training, advertising, and data monetization; requiring 24-hour breach notification; and demanding independent efficacy…
Stanford found strong evidence for human-led tutoring, emerging evidence for AI supporting tutors, and insufficient evidence for fully automated tutoring. AI recommendations increased lesson-topic mastery by four percentage points overall and nine points with…
Early field use of CYBER.ORG’s nine caregiver lessons shows schools distributing AI and online-safety guidance through counselors, newsletters, and school apps. Modules cover three age bands.
Iowa City launched a formal review of screen time, take-home devices, instructional value, privacy, accessibility, and one-to-one program outcomes amid parent concern and cost pressure.
Meta agreed to pay up to $18 billion and limit teen Facebook and Instagram use, including daily limits, overnight restrictions, and reduced school-hours notifications. Claims included children’s data being used for generative AI training.
EDSAFE AI selected ten organizations for $250,000 in grants and technical assistance. Norwalk Public Schools received $23,000 to address privacy, procurement, academic integrity, staff literacy, and parent engagement.
Canvas is testing ways to show how students used AI during assignments. Anthropic is adding text watermarking and plans a detection interface, but acknowledges that watermarking indicates probability, not authorship.
On September 9, Microsoft, the AFT and UFT announced a standard that districts can incorporate into Microsoft customer agreements. It prohibits using student or educator data to train AI, selling or repurposing that data, student tracking and unsupervised…
An outside group accessed Springfield Public Schools’ network and blocked essential online programs. The district classified it as a Level 4 severe incident and closed schools Tuesday through Friday because nurses could not access medication and allergy…
Governor Newsom signed a 13-bill child-technology package on September 10. It includes independent safety assessments and parental controls for companion chatbots, restrictions on addictive feeds for children, and stronger pupil-data protections. AB 1159…
More than 600 participants joined an AASA town hall on September 10 as the FCC considers narrowing, geographically limiting, restructuring or potentially ending parts of E-rate. The program currently provides 20%–90% discounts. The FCC’s 2026 funding cap is…
Missouri and Google announced no-cost enterprise AI tools, training and credentials for nearly 100,000 educators and more than 1.1 million K–12 and higher-education students. Offerings include Gemini for Education, Gemini Notebook and professional…
State dispute-resolution agencies report receiving lengthy filings containing incorrect IDEA interpretations, erroneous filing requirements and hallucinated case law. Evidence linking AI to overall complaint growth remains anecdotal. Separately, written state…
Mathspace disclosed that attackers exploited an unpatched vulnerability in its self-hosted Metabase reporting environment. The flaw was patched by Metabase on August 6; unauthorized access began August 10, while Mathspace updated its instance August 29.…
For 108,000 students, Baltimore County reported average daily school-device use of 70–90 minutes in middle school and 75–100 minutes in high school. Grades 3–5 averaged 50–60 minutes; K–2 averaged about 30 minutes. A separate 344-district benchmark covering…
Jordan School District and SchoolAI reported a 28% increase in a rubric-based measure of critical thinking across 13,882 student-AI conversations involving 82 teachers over two years. Higher-level reasoning classifications more than doubled, with stronger…
OECD reading scores fell 14 points from 2022. PISA links AI use for summarizing, drafting, and preliminary research with lower science scores. Weekly use for general learning shows similar scores to non-use; opportunities to assess AI information accompany…
Teacher candidates using chatbot simulations improved more in lesson planning, pedagogical knowledge, and teaching confidence than the comparison group. The content-knowledge difference was not statistically significant.
Participants preferred plans and decomposed explanations, but simpler step-by-step traces often better supported verification. Some preferred formats increased false alarms or encouraged trust without corresponding willingness to check.
Students reported greater ethical awareness and reasoning confidence after discussions with AI personas and peers. Interviews also identified overly agreeable AI responses and limited interaction between human participants.
Sustained social engagement with Character.AI was associated with lower well-being. Less in-person interaction statistically explained much of that relationship.
Clinician raters identified descriptions consistent with delusional beliefs in 102 of 185 harm reports; 50 of those described chatbot validation. Reported consequences included isolation, relationship breakdown, and hospital admission.
UNESCO frames AI adoption around student development, autonomy, pedagogy, assessment, expression, and public accountability. It calls on education systems to actively shape their response to algorithmic technologies.
Students were more positive about AI than staff, while staff expertise and perceived uses varied. Student uses centered on language improvement and text production; appropriate academic use remained ambiguous.
A smart-kitchen-appliance analogy illustrates how AI can automate an outcome or help a learner practice. The authors argue that the mode of use and resulting cognitive activity determine its educational value.
The hypothesized relationship between frequent AI use and lower performance self-esteem was not supported. The observed correlation was near zero (r = -0.039; p = 0.675).