Edition 23 | June 13, 2026 · Matchday
This week's theme: kickoff, and a red card. The World Cup opened on home soil this week, and AI had a matchday of its own. On Tuesday, Anthropic put the most powerful model it has ever released to the public, Claude Fable 5, on the field. By Friday the US government had ordered it pulled, three days after launch. Between those whistles, the week showed how deep AI has already moved into schools: most kids now use it, the evidence for AI tutoring keeps building, special-education teachers are leaning on it to write IEPs, and the largest district in the country faced a public call to pause it altogether. The ground under the tools shifted too, with the largest education data breach on record still coming into focus. A week of dramatic goals and sudden red cards, with plenty for school leaders to take back to their own bench.
On June 9, Anthropic released Claude Fable 5, the first publicly available version of its most capable model line, alongside a less restricted variant called Mythos 5 for vetted partners, and called Fable 5 its most powerful public model to date. To manage risk, Fable 5 declines to answer in high-risk areas such as cybersecurity, biology, and chemistry, handing those questions to the steadier Claude Opus 4.8 instead, and Anthropic says about 95 percent of sessions run entirely on Fable without that handoff. It launched free inside Pro, Max, Team, and Enterprise plans through June 22, after which access requires usage credits. Pricing runs about 10 dollars per million input tokens and 50 dollars per million output tokens, roughly double the cost of Opus 4.8.
Why it matters: The frontier of what these tools can do moved again this week, and the gap between the free model your students reach for and the priced flagship behind it keeps widening. For school leaders the lesson is less about any single model and more about pace, because capability is now shipping faster than policy can absorb it, which is exactly why a standing, model-agnostic AI policy beats chasing each release. Worth noting too: the safety design here, refusing high-risk domains by default, is the kind of guardrail worth asking any classroom vendor to document.
A new Common Sense Media report finds that 86 percent of young people have used generative AI tools such as ChatGPT and Google Gemini, with use rising by age, from 81 percent of 9 to 12 year olds to 92 percent of 16 and 17 year olds, and nearly a quarter using AI every day. Most reach for it to have fun, but a majority also use it for schoolwork, and 57 percent have asked it for advice about their health or their bodies. The gap the report highlights is guidance: 44 percent of kids say they have never talked about AI safety with a parent, and only about a third understood that AI cannot reliably tell fact from fiction.
Why it matters: This is the adoption curve underneath every policy debate. Students are already using these tools, daily and largely unsupervised, including for sensitive questions a search engine never answered with such confidence. The takeaway is that bans do not describe reality and silence is not neutral, because if school and home are not teaching how AI gets things wrong, kids form those habits without that frame. Pair any device or access rule with explicit instruction on checking AI output, and give families a short, plain-language guide so the conversation happens somewhere.
On June 12, a majority of the New York City Council, 29 members, sent a letter urging Mayor Zohran Mamdani and schools Chancellor Kamar Samuels to pause the use of artificial intelligence in New York City Public Schools, the largest school system in the country, for two years. Drafted by Councilwoman Alexa Aviles with education advocates, the letter calls the city Department of Education's draft AI guidance flawed, faulting it for not strengthening student-data privacy protections around AI companies, and raises concerns about children's cognitive development, creativity, mental health, and the environmental cost of AI. It points to an April audit by the New York State Comptroller that flagged compliance problems. The schools department said it would share more about its AI policy soon, adding that the prior administration moved on AI without genuine family engagement.
Why it matters: When the nation's largest district faces a public call to hit pause, it signals that AI in schools is now a political and community question, not just a technical one. Whatever your view on a moratorium, the council's specific objections are a useful checklist: can you show families exactly how student data is protected when an AI vendor is involved, and did you engage parents before rolling a tool out, not after. Districts that can answer those two questions calmly will weather this scrutiny. Those that cannot should expect a version of this letter to arrive locally.
A FutureEd research note from Georgetown gathers early evidence that well-designed AI tutoring can help students learn. In a study of Google's LearnLM with the platform Eedi, supervising tutors approved about 76 percent of the model's responses with little or no editing, and students who worked with LearnLM reached a 66 percent success rate on later challenging topics, compared with 61 percent for human tutors and 56 percent for static hints. A separate Stanford study of Tutor CoPilot, which coaches human tutors in real time, found students were 4 percentage points more likely to master a topic, rising to 9 points when their tutor was lower rated, and tutors using it were far likelier to ask students to explain their thinking.
Why it matters: This is the hopeful counterweight to the shortcut worries. Used as a tutor, or as a coach for tutors, AI is starting to show measurable gains, especially for the students and staff who need the most support. The design detail matters more than the headline, because the wins came from tools built to prompt reasoning and keep a human in the loop, not from a chatbot handing over answers. When you evaluate an AI tutoring product, ask for trial evidence, and look at whether it pushes students to explain their thinking or simply finishes the task for them.
An NPR report this spring put numbers to a fast-growing practice. According to a Center for Democracy and Technology survey, 57 percent of special-education teachers used AI to help develop IEPs in the 2024 to 2025 school year, up from 39 percent the year before, and 15 percent said they relied on AI entirely to develop those plans. The pull is understandable, because 45 states reported special-education teacher shortages and the paperwork is crushing. Teachers profiled in Mt. Diablo Unified in California described using AI to save time so they could spend more of it with students. But the report's lead author at the Center for Democracy and Technology, Ariana Aboulafia, and the teachers themselves stress the risk. As one put it, the key word is individualized, and no two children are the same.
Why it matters: An IEP is a legal document that must be individualized to one child under the Individuals with Disabilities Education Act, and a plan drafted largely by AI can fail that standard even when it reads well. There is a privacy risk too, because pasting a student's details into a consumer chatbot can expose protected information. The move is not to ban the tool but to govern it: tell special-education staff which approved, privacy-safe tools they may use, require that a human author owns and tailors every plan, and make clear that identifiable student data never goes into a personal AI account.
The scale of a data breach at Instructure, the company behind the Canvas learning platform, came into fuller view this spring. The cybercrime group ShinyHunters claimed to have taken roughly 3.65 terabytes of data tied to about 275 million users across thousands of institutions, reportedly by abusing Canvas's free teacher accounts, with private student and teacher messages among the exposed data. Security analysts have described it as the largest breach in the education sector to date. Some of the biggest figures are the attackers' own claims rather than confirmed counts, and the investigation continues, but the event has already become a reference point for edtech risk.
Why it matters: The biggest threat to your students' data may not be your own network, it is the vendors who hold that data for you. A breach at a single platform can expose tens of millions of records at once, far beyond what any one district could leak on its own. Treat vendor security as a purchasing decision, not an afterthought: ask where data lives, how access is controlled, and how fast a breach gets disclosed, and write those answers into the contract. The free tier that is easiest to adopt can also be the softest target.
Stepping back from this week for the bigger picture: in late May, the California State University system, the largest public university system in the country, renewed its systemwide ChatGPT Edu contract with OpenAI at about 13 million dollars a year for three years, down from 17 million for the previous shorter deal. The agreement covers more than 470,000 students and 63,000 faculty and staff across 22 campuses, and OpenAI calls it its largest higher-education partnership. The move is not without friction: a faculty petition earlier in the year argued the tool was not designed or optimized for education, a campus survey found 84 percent of students already use ChatGPT while 82 percent worry about its effect on job security, and the system placed the bet while facing 144 million dollars in budget cuts.
Why it matters: K-12 leaders should watch higher education here, because the district-wide AI license is coming for you next, and CSU is the template vendors will cite. The lessons travel down. Notice that even a flagship deal drew faculty pushback over whether a general consumer tool fits real academic work, that students were confused by inconsistent classroom rules, and that the spend landed in a tight budget year. Before you sign anything systemwide, define the instructional purpose first, write in data protections and an exit, and set one consistent use policy so teachers and students are not left guessing.
Apple held its Worldwide Developers Conference from June 8 to 12, with the keynote on June 8. The headline artificial-intelligence news was aimed at consumers rather than classrooms, so the school story is quieter and worth separating from the hype. According to an education recap by the device-management firm Jamf, the updates that matter most to schools center on management and security: improvements to device management, identity and single sign-on, credential and password syncing, expanded and offline authentication, and better status reporting for the administrators who run fleets of school devices. Apple also spotlighted students, naming 350 Swift Student Challenge winners, 50 of them invited to Cupertino.
Why it matters: It is easy to read every Apple AI announcement as a classroom story, and this one mostly is not, so telling the difference is part of the job. The real wins for schools this year sit in the unglamorous layer your technology team lives in: smoother device management, stronger identity and authentication, cleaner reporting. Those changes reduce friction and shore up security, which matters more day to day than another headline feature. Ask your technology lead which of these updates affect your devices before the new school year, and let the consumer AI news stay in its lane.
Two educator-focused gatherings anchored the week. The 1EdTech Engage Learning Conference ran June 9 and 10, hosted by Mountain Brook Schools in Alabama, with free registration and a practitioner audience focused on putting edtech to work in real classrooms. Right behind it, the University of Central Florida hosted Teaching and Learning With AI, a sharing conference running June 11 through 13 at the Gaylord Palms resort in Orlando, organized by UCF's Faculty Center for Teaching and Learning, its Division of Digital Learning, and UCF Libraries. Both reflect how much of the practical AI conversation is now happening teacher to teacher, in sessions built around what actually works.
Why it matters: The most useful AI professional development right now is often peer to peer, and conferences like these are where classroom-tested ideas spread faster than any vendor pitch. You do not need a travel budget to benefit, because many sessions post slides, recordings, and resource lists afterward, so it is worth assigning someone to follow along and bring three usable takeaways back to your team. If you are planning your own professional development, the format here, short sharing sessions over keynote lectures, is a model worth borrowing.
On June 12, three days after Fable 5 went public, Anthropic disabled both Fable 5 and Mythos 5 for all users to comply with a US government directive. According to the company's own statement, the government issued an export-control order to suspend access to the models by any foreign national, anywhere, including Anthropic's own foreign-national employees. Because Anthropic could not separate foreign nationals from US users in real time, it switched the models off for everyone. All of its other models, including Claude Opus 4.8, stayed available. The stated reason was a claimed method of jailbreaking Fable 5, which Anthropic says relates to a narrow technique already available in other models, and it says it received only verbal evidence. News coverage reported the order came in a letter from Commerce Secretary Howard Lutnick to chief executive Dario Amodei.
Why it matters: This is the week's hardest lesson for anyone building on AI: access can vanish overnight, by order, not just by outage. A tool your staff or students start to depend on can be switched off for reasons that have nothing to do with you, and with no notice. That is a planning fact, not a reason to panic, so never let a single model become a single point of failure in a lesson, a workflow, or a contract. Build in a fallback, keep your own copies of anything that matters, and favor vendors and plans that let you move. Capability is real, and so is its fragility.
Try This Week
Run a 15-minute matchday review with your team. This week a powerful tool launched and was pulled within three days, so open with the single-point-of-failure question: if an AI tool your school depends on went dark tomorrow, by outage or by order, what breaks, and what is your fallback? Pick the three you would miss most and name a substitute for each. Then play some defense, because the breach and the finding that most kids already use AI on their own both point to one question: who in your building can explain, today, exactly what student data leaves it and how families are told? Finally, read the home crowd, since the largest district in the country is being asked to pause AI over privacy and engagement: could you show a parent your data protections and your family-engagement steps on a single page? You do not need every answer this week. You do need to know which one would cost you the match.
Until next time,
Dr. Janette Camacho
CEO, iTeachAI Academy
Your goal this season? Renew your certification. iTeachAI Academy has state-aligned professional development that counts toward recertification in all 50 states, with a built-in AI guide in every course. Don't let your hours run down the clock. See what counts in your state → |
Free AI courses at classes.iteachai.co
17 free AI tools at iteachai.co/TeacherTools
Know a teacher who needs this? Forward this email.
Subscribe free at iteachaibot.com
The top AI stories, tools, and practical tips for your classroom. Written by educators, for educators.
Join 1,400+ educators. Unsubscribe anytime.