Everyone selling certainty got asked for evidence this week. The receipts came in.
This was the week education asked for receipts. A child safety lab audited the AI search features sitting on every school Chromebook and failed them. Hawaii wrote the first real rules for the companion chatbots kids use after school, and Illinois became the first state to say an algorithm may not judge a teacher. The first big randomized trial of teacher facing AI found no learning gains and less motivated students. Ten million students have nine days left to claim their share of a privacy settlement, and a government audit of an entire edtech market concluded the evidence mostly comes from the people selling the tools. None of this says AI has no place in your classroom. It says the burden of proof just moved to the sellers, where it always belonged. Here is what happened, and what to do with it.
Google's AI Search Just Failed Its Child Safety Audit
Common Sense Media's Youth AI Safety Institute rated Google Search's AI Overviews and AI Mode an overall Unacceptable Risk for children and teens. Researchers ran more than 2,600 interactions on accounts set up as an 11 year old and a 15 year old, with SafeSearch on, and reviewed over 2,100 of the sources the AI cited. The findings: appropriate crisis resources surfaced in barely more than half of clear crisis moments, eating disorder prompts got hotline referrals only 38 percent of the time, both features handed out step by step deepfake tool recommendations, and more than a quarter of citations came from forums and social posts dressed up as authority. On the academic side, AI Mode produced a complete, submittable answer for 100 percent of the 180 assignments tested. Google disputed the methodology, calling the queries contrived.
Why it matters: This is not an optional chatbot a district can decline to buy. AI Overviews appear automatically on the same school Chromebooks students use every day, and neither schools nor families can switch them off separately from Search. Assume every take home task is one search away from done, teach source verification explicitly, and loop in your counselors on the crisis response gaps: the tool most students touch first is the one nobody vetted.
Hawaii Just Regulated the AI Kids Actually Use After School
Governor Josh Green signed two laws on July 14 aimed at the AI in students' pockets. The first covers companion chatbots: operators must clearly disclose that users are talking to an AI, repeat that disclosure regularly in long conversations with minors, build crisis protocols that route users expressing self harm toward real help, stop bots from posing as licensed mental health professionals, and avoid addictive design aimed at children. The second creates deepfake protections, letting victims of harmful AI digital imitations sue for damages up to 25,000 dollars per piece of content. Green said the measures were inspired partly by reports of young people receiving dangerous advice from chatbots. Rhode Island signed similar chatbot measures the same week.
Why it matters: Companion chatbots are where many students actually spend their AI hours, long after the school Chromebook closes. Hawaii's rules read like a checklist your advisory or digital citizenship lesson can borrow today: does the bot say it is a bot, what does it do when someone is struggling, and who is it pretending to be. Expect versions of this law in more statehouses within the year.
First Big Randomized Trial of Teacher AI: No Gains, Less Motivation
A working paper from Wharton's Alp Sungu with Benjamin Lira and Angela Duckworth ran the first large randomized trial of a teacher facing AI assistant in real classrooms: 193 teachers, 2,816 students, 14 schools in Turkey, a GPT-4o assistant loaded with the national curriculum. The result: no average improvement in grades, a measurable decline in students' intrinsic motivation, and students of initially weaker teachers actually scored lower on standardized finals. Two thirds of teacher usage was generating materials; almost none was deepening feedback. The paper is a draft and not yet peer reviewed.
Why it matters: This is a direct hit on the industry's favorite syllogism, that faster teachers automatically mean better learning. Students noticed AI assisted teaching and rated those classes less interesting and less important. The lesson is not to ban the tools; it is that time saved on materials only pays off if you reinvest it in the human work, the feedback, the conferences, the questions. AI that replaces preparation can quietly replace the thinking that made preparation valuable.
Claude for Teachers Goes Free as the Labs Compete for Your Classroom
Anthropic announced Claude for Teachers on July 14: free for verified US K-12 educators, built around academic standards from all 50 states, and aimed at lesson planning, differentiating materials, and analyzing classroom data. Two details stand out from the coverage. The privacy posture was negotiated with the American Federation of Teachers, and model training is switched off on verified teacher accounts. And the timing was no accident: the AFT's national convention opened two days later, and every major lab now has a free teacher offering in market. Students under 18 still cannot use Claude directly. Education Week's survey data shows why the labs are competing: teacher AI use jumped from 32 percent in 2024 to 61 percent in 2025.
Why it matters: Free is a price, not an evaluation. This one arrives with two things most classroom AI tools still lack, a union negotiated privacy standard and training switched off by default, and those are exactly the receipts to ask every other vendor for. If you claim a free teacher account this week, from any lab, run it through the same three questions as a paid tool: what evidence, whose data, and who is accountable when it is wrong.
Naviance Settlement: 10 Million Students Can Still File Until July 27
The proposed 17.25 million dollar settlement over PowerSchool's Naviance college and career platform is in its final stretch. The suit alleged that analytics and advertising technologies embedded in Naviance intercepted students' communications and sensitive information without proper consent; PowerSchool, Hobsons, analytics firm Heap, and the Chicago Board of Education all deny wrongdoing. The class covers roughly 10 million people who logged into Naviance between August 2021 and January 2026 across some 13,000 schools. The deadline to object passed on July 13; the deadline that still matters is July 27, the last day class members can file a claim. The agreement also forces governance changes: vetting embedded analytics, clearer disclosures, and honoring deletion requests.
Why it matters: Naviance runs in thousands of high schools, which means many readers of this newsletter, your students, your own kids, possibly you, are class members with nine days to file. Beyond the money, the governance terms are a ready made checklist for your next procurement conversation: what analytics ride inside this platform, who else receives the data, and can a family actually get it deleted.
A Texas Career College Breach Shows What Schools Really Hold
The College of Health Care Professions, a Houston based allied health college with nine Texas campuses, confirmed that attackers spent four days inside its network in August 2025. By the college's own report to the Texas Attorney General, files for 68,825 people may have included Social Security numbers, government IDs, financial account details, medical information, and health insurance records. Notification letters did not go out until late February, roughly six months after the intrusion. The Houston Chronicle reported this week that the case has now drawn a consolidated class action petition; multiple law firms had already opened investigations. Those are allegations, not findings, and the college has not commented publicly.
Why it matters: Career colleges concentrate an unusual cocktail of identity, financial, medical, and education data, but so do K-12 districts, and so do your vendors. The two questions this case hands every school leader: exactly what data do we and our platforms retain about students and families, and how fast would we tell people if it leaked. Six months of silence is the reputational injury on top of the breach.
One Cloud Hiccup Took Down Canvas Worldwide for Four Hours
On July 16, teachers trying to reach Canvas, Catalog, Mastery Connect, or Portfolio hit errors for about four hours. The culprit was not Instructure at all: a global failure in Amazon's CloudFront network knocked out services across the internet, Blackboard included. Instructure's status page tracked the recovery region by region, and a separate, smaller incident the same day briefly broke the AI tools inside Canvas. Both were resolved by day's end, and no data breach was involved.
Why it matters: Two takeaways worth filing before fall. First, your LMS now sits on shared cloud plumbing, so one upstream failure can take out your platform, your backup platform, and the status pages in between; keep a low tech plan for lesson one. Second, notice that the AI features failed separately from the platform itself. As AI rides into every tool you use, you inherit its failure points too, and a bookmark to status.instructure.com beats blaming your own IT at 7 AM.
Illinois Is First to Ban AI From Judging Teachers
Governor JB Pritzker signed SB 2909, making Illinois the first state to prohibit evaluators from using AI to assign any score or rating in a public school teacher's evaluation. The law, which passed both chambers unanimously and takes effect January 1, 2027, also bars teachers from using AI to write their own portion of an evaluation. Sponsor Senator Christopher Belt put it plainly: teachers should be judged on actual observations and professional judgment, not by AI software.
Why it matters: A unanimous vote is a receipt of its own: across parties, lawmakers agreed that high stakes judgments about educators belong to humans. It is the same principle this newsletter keeps finding in the classroom research, judgment is the part AI cannot carry. Watch for copycat bills, and note the symmetry: if the algorithm cannot judge your teaching, it cannot write your self evaluation either.
Baltimore County Parents Pushed, and Screen Time Policy Is Coming
At its July 14 meeting, the Baltimore County school board unanimously backed drafting a policy on student screen time after parents described Chromebooks open on desks most of the day, students racing through assignments to earn game time on Prodigy, reading on Epic mainly to collect badges, and one parent whose daughter's AI reading program could not understand her speech. The coming standards would cover instructional versus noninstructional use, age appropriate limits, accommodations for students with disabilities, and communication with families.
Why it matters: The edtech question in your community is shifting from is this app safe to how much of the school day should run through a screen at all, and gamified rewards are exhibit A. Teachers should expect to justify device time as instructional, not filler, and the speech recognition complaint is its own quiet warning: an AI tool that cannot understand a child is not neutral, it is a barrier wearing a progress bar.
The UK Audited Its Whole Edtech Market. The Verdict Stings Everywhere.
A 104 page assessment of England's edtech market, commissioned by the Department for Education and published in late June, keeps drawing sector attention, and it reads like a diagnosis of the global market, not just Britain's. Among its findings: general purpose AI tools were not built with school specific safeguarding as a primary focus, platform lock in is real as switching costs pile up, the evidence base is thin and often vendor produced, long term evaluations barely exist, and schools are being asked to judge privacy clauses, data sharing, and algorithmic bias with expertise nobody gave them.
Why it matters: Swap England for any US state and every sentence still lands. The structural problem this week's stories share is right here: schools carry the consequences of tools they lack the evidence and leverage to evaluate. Until independent evidence catches up, the practical move is borrowed diligence: demand independent research rather than vendor case studies, ask what happens to your data at contract end, and treat every AI feature as unproven until someone outside the company says otherwise.
Try This Week
Three moves, thirty minutes total. First, the receipt check: if your school used Naviance anytime between August 2021 and January 2026, you and your students may be class members; filing a claim takes five minutes at the settlement site above and the door closes July 27. Second, the vendor test, borrowed from California's audit: pick one AI tool your school pays for and ask the vendor three questions in one email: what independent evidence, not your own case studies, shows this improves learning; what happens to student data at contract end; and what are the performance benchmarks in our agreement. The quality of the answer is the answer. Third, the September check: pick one AI tool you plan to lean on this fall and hold it up against your district policy line by line. If a claim and a clause disagree, you have found your first question for the vendor, and your first conversation for the staff room.
Every state-aligned iTeachAI Academy course counts toward your recertification, comes with a built-in AI guide that does the heavy lifting, and takes about as long as a good episode of your show. AI integration, structured literacy, MTSS, SEL, and your state's mandated topics, all online, all on your schedule.
One course, $25. All-access, all year, just $149.
See what counts in your state →Professional development that respects your weekend.
Receipts season is good for education. The more proof we demand from the tools, the better the tools get, and the more your professional judgment is worth. Keep asking.
Until next time,
Dr. Janette Camacho
CEO, iTeachAI Academy
P.S. New this summer: iTeachAI Academy Global. Teacher recertification tracks mapped to each country's own standards and licensing rather than generic content, offered in English and the local language. The United Arab Emirates, Qatar, Saudi Arabia, England, Canada, and Brazil are enrolling now, with more on the way. Explore the world catalog at classes.iteachai.co/world.