{"slug": "google-gemini-in-classrooms-who-consents-for-an-eleven-year-old", "title": "Google Gemini in Classrooms: Who Consents for an Eleven-Year-Old?", "summary": "Google began switching Gemini on by default in Google Classroom for students of all ages on 10 August 2026, with mobile following on 17 August, removing an age gate that had limited automatic Gemini access to students 18 and over on a platform used by more than 150 million students and teachers worldwide. On Tuesday 3 November, Google vice president and general manager for education Chris Phillips announced that Gemini, Gemini Notebook, Gemini in Classroom and Google Search would be \"entirely off for K-12 student accounts until school administrators explicitly turn them on,\" reversing the August default 85 days later. The episode drew criticism from district technology leaders including Palo Alto Unified School District CTO Derek Moore, who said \"It feels like we're the last to know,\" and came after Anthropic launched Claude for Teachers on 14 July, giving verified US K-12 educators free access to its premium tier and standards-mapped lesson planning, with a full year at no cost for signups before 30 June 2027.", "body_md": "## Google Gemini in Classrooms: Who Consents for an Eleven-Year-Old?\n\nThe most consequential sentence written about American education in August 2026 ran to eleven words and appeared in a Google product bulletin, under a subheading called “Rollout pace”. It reads: “On by default for teachers and students of all ages.”\n\nThat is the whole story, compressed. Not a policy paper, not a hearing, not a school board vote. A default.\n\nOn 10 August 2026, Google began switching Gemini on inside Google Classroom for students of every age on the web, with mobile following on 17 August. Classroom is used by more than 150 million students and teachers worldwide. Until that week, Google automatically enabled Gemini in Classroom only for students aged eighteen and over. On 14 August the New York Times reported the change, and the reporting travelled fast, because the timing was exquisite. American children were walking back into buildings they had left in June, opening the same application they had used all of the previous year, and finding a new tab.\n\nDerek Moore, chief technology officer at Palo Alto Unified School District, described the experience of Google's cadence of AI changes in one line: “It feels like we're the last to know.” James O'Hagan, director of digital equity, innovation and library services at the Glendale-River Hills School District in Wisconsin, put the principle more bluntly: “The default should be requiring permission, not automatic adoption.”\n\nEighty-five days later, Google appeared to agree. On Tuesday 3 November, Chris Phillips, Google's vice president and general manager for education, announced that Gemini, Gemini Notebook, Gemini in Classroom and Google Search would be “entirely off for K-12 student accounts until school administrators explicitly turn them on”. Search was on that list, and that is no accident. The default for millions of children had now moved twice in a single term, once by product bulletin and once by blog post, in opposite directions. Neither move went through a parent.\n\nFour weeks before the August switch, on 14 July, Anthropic had launched Claude for Teachers, giving verified American K-12 educators free access to its premium tier, a library of teaching skills, and lesson planning grounded in curricula mapped to academic standards in all fifty states. Educators who sign up before 30 June 2027 get a full year at no cost.\n\nSo within a month, the two most heavily capitalised AI laboratories in the United States both moved decisively into the K-12 classroom. One did it by flipping a switch on a platform it already owned. The other did it by handing a licence to a teacher. Those are not the same act, and the difference is where the entire accountability question lives.\n\n## What Changed in August, What Changed Back in November, and What the Headlines Missed\n\nPrecision first, because the loose version of this story is doing rounds and it is wrong in a way that matters.\n\nGoogle did not reach into every school in America and turn on a chatbot for every child. The August change applies to students who had already been granted access to Gemini in Classroom, the Gemini app or Gemini Notebook by their district administrator. Where a district had switched Gemini off for students, Google's own documentation states it “will remain off for all students, including students younger than 18”. Administrators retain the ability to carve out organisational units and disable access for under-eighteens specifically.\n\nWhat changed, then, is narrower and stranger than “Google turned on AI for children”. What changed is that an age gate came down. Districts that had enabled Gemini for students on the reasonable understanding that this meant their sixth formers, their eighteen-year-olds, the students Google had itself defined as the eligible population, woke up to find the same permission now flowing down to eleven-year-olds. The consent the district gave in January was for one population. In August it silently covered another.\n\nThis is a familiar manoeuvre in software, and it has a name in privacy circles: scope creep by default. The setting did not change. The meaning of the setting changed.\n\nThe Youth AI Safety Institute at Common Sense Media, which assessed Gemini K-12 in January 2026 and rated it High Risk overall, made the design consequence explicit. The product treats a sixth grader and a high school senior the same. Guided Learning mode, Google's pedagogical guardrail, is easy to switch out of, after which the model will write a full essay. Common Sense's separate assessment of AI Overviews and AI Mode in Google Search, which no user in a school can disable, rated them Unacceptable Risk after the system completed one hundred per cent of one hundred and eighty tested homework assignments.\n\nA further default has since landed on a different data type. Google is rolling out an opt-out setting called “Personalized Recommendations and Search Services History” to all accounts, under-eighteen school accounts included, which tailors what students see using their saved activity and location data. Google's Workspace for Education documentation gives administrators a switch to turn personalised recommendations off for users designated as under eighteen, and notes that doing so neither turns off the history setting nor deletes activity already saved. Same architecture. Same default posture. The burden of noticing sits, once again, with the administrator.\n\nThen, on 3 November, the default moved back. In a post on Google's Keyword blog titled “Designing responsible AI for learning”, Phillips said Gemini, Gemini Notebook, Gemini in Classroom and Google Search would all start switched off for K-12 student accounts, with schools able to enable each individually. “We believe schools should have control over AI in their classrooms,” he wrote on X.\n\nThat sounds like what O'Hagan asked for in August, and in part it is. Look at the last item on the list, though. Google is building Guided Learning into Search for student accounts, with a pilot in select American schools later in November, and Search now comes with the AI attached. Phillips said schools that ban AI “will not have access to search”. Later that day Google clarified that schools could enable Search without Gemini, but Search would still carry AI Overviews and other AI features. The company said the announcements “give schools more choice, not less”.\n\nThe reaction was swift. New York City Mayor Zohran Mamdani accused Google of “trying to use its monopoly power to force AI into our public schools”. Lina Khan, the former chair of the Federal Trade Commission, said the move “highlights the danger of allowing an unregulated monopolist to control an essential service”. Josh Golin, executive director of Fairplay, said Google's effort to “strong-arm schools into using its AI products is likely to backfire”. Randi Weingarten, president of the American Federation of Teachers, said: “It is unfortunate that Google has prioritized its business interests over the education of our children.”\n\nSome districts had already stopped waiting for Google. On 2 September, New York City announced a one-year pause on student-facing generative AI for students through eighth grade, with companion chatbots prohibited at every grade. Los Angeles Unified temporarily blocked generative AI tools on student devices across all grades while it writes a policy.\n\nNone of Google's November changes has taken effect. The company says it will “share more specifics with schools in the coming months ahead of the changes rolling out”. So the August question, whether consent given for eighteen-year-olds silently covered eleven-year-olds, has been answered with a different lever. The new question is whether a district can keep a search engine without taking the AI built into it.\n\n## The Four Doors a Decision Walks Through Before It Reaches a Parent\n\nUtah is the cleanest available illustration of how consent evaporates as it descends.\n\nOn 2 June 2026 the Utah State Board of Education announced the launch of Gemini for Education across the state's public schools from the 2026-27 academic year, in partnership with Google. The scale is roughly 680,000 K-12 students and 28,000 educators. The press release is careful, and to its credit it is careful in the right direction: local education agencies “maintain full discretion over whether to implement Google Gemini for Education”. State Superintendent of Public Instruction Molly Hart framed it around local control. Matt Winters, the board's AI specialist and, since 2024, the first full-time state-level position of its kind in the country, tied the programme to House Bill 273 and to helping districts screen software for security and privacy.\n\nRead the chain of authority carefully.\n\nThe state board negotiates the agreement. The local education agency decides whether to adopt. The district administrator configures the console. Google sets the default inside the console. And then, in August, Google changes what that default covers, and in November announces it will change the default again on its own terms.\n\nAt no point in that chain does a parent appear. At no point does a student appear. Four institutional actors each exercised legitimate discretion, each did so within the bounds of their authority, and the cumulative result is a thirteen-year-old typing into a large language model without anybody having asked either of the two people whose data it is.\n\nThis is not a conspiracy. It is worse than a conspiracy, because it is nobody's fault in particular and therefore nobody's problem to fix. Every actor in the chain can point upstream or downstream. The state points at local control. The district points at the state contract. The administrator points at the console default. Google points at the administrator.\n\nThe Utah release does contain a genuine substantive commitment, and it deserves to be stated plainly rather than waved away: interactions remain within a secure domain, data and conversations are not used to train AI models, and institutions retain administrative control over their data. Google's Gemini for Education materials say the same, adding that data is not human-reviewed and not used for ad targeting, and listing COPPA and FERPA among the frameworks supported.\n\nTake those commitments at face value. They are probably accurate. And notice that they answer a question nobody in this story is actually asking.\n\n## Why the Strongest Case for Doing This Deserves a Proper Hearing\n\nThe lazy version of this article treats free AI in schools as obvious predation. That version is wrong, and it is wrong in ways that make the real critique weaker.\n\nStart with the labour arithmetic. RAND's 2025 State of the American Teacher survey found teachers working an average of forty-nine hours a week. Gallup research with the Walton Family Foundation found that teachers who use AI tools at least weekly estimate savings of 5.9 hours per week, which across a 37.4-week school year amounts to roughly six weeks of recovered time. Three in ten teachers were using AI weekly.\n\nSix weeks. For a profession that has spent two decades being told its workload problem is a personal resilience problem, that is not a rounding error. It is the difference between a teacher who plans a lesson properly on Sunday evening and a teacher who resigns in March.\n\nThen the equity case, which is the strongest one and the least discussed. Affluent families already have AI in the home. They have subscriptions, they have parents who know what a system prompt is, and they have children who were using these models within a fortnight of release. If the tools stay out of school, they do not stay out of childhood. They simply become another axis along which advantage compounds quietly. A free, district-managed, contractually constrained deployment inside a school is, on this reading, the only version of AI access that reaches the child whose household has no subscription and no fluent adult.\n\nThird, the budget reality. Education's AI sector reached roughly 730 million dollars in 2026 and is projected to grow around forty per cent annually for the next decade, reaching an estimated 18.5 billion dollars by 2036. Districts are not swimming in money. When a vendor offers a capability at zero marginal cost against a paid alternative, the procurement conversation is effectively over before it starts.\n\nAnd Anthropic's offering, specifically, is built to answer the pedagogical objection rather than dodge it. Claude for Teachers draws standards data from Learning Commons, a non-profit database covering all fifty states, and pulls curricular material from OpenSciEd and Illustrative Mathematics. It connects to nine existing K-12 platforms including ASSISTments, Brisk Teaching, Canva Education, Diffit, MagicSchool and TeachFX. It ships an AI Fluency course that is model-agnostic and Creative Commons licensed, which is to say Anthropic wrote training material that teaches teachers to use its competitors. Its partners include the American Federation of Teachers and Teach For America.\n\nNone of that is nothing. A company that wanted only capture would not release its literacy curriculum under a licence permitting anyone to strip its branding.\n\n## What Free Actually Buys and Why the Usual Slogan Fails Here\n\n“If you're not paying, you're the product” is the reflex, and in this instance the reflex is close to useless.\n\nUnder FERPA's school official exception, a vendor handling education records must perform a function the institution would otherwise perform with employees, operate under the direct control of the institution regarding use and maintenance of those records, use personally identifiable information only for the purpose of the disclosure, and satisfy the criteria set out in the school's annual FERPA notification. Layer on state law and the constraint tightens further. California's SOPIPA, passed in 2014 and effective from 2016, bars targeted advertising, commercial profiling and the sale of covered student data, and it binds the vendor whether or not a contract exists. Illinois SOPPA, effective July 2021, requires published vendor lists and deletion within sixty days of a school's request. New York Education Law 2-d requires a Parents' Bill of Rights and a data security plan in every contract.\n\nAgainst that stack, the crude surveillance-capitalism reading genuinely does not hold. Google says school Gemini data is not used to train models or serve ads. Anthropic says Claude for Teachers data is not used for model training and that student information is covered by a K-12 Data Processing Addendum written to comply with FERPA. Both statements are contractual commitments with legal exposure attached.\n\nSo what is the currency?\n\nIt is not data. It is dependency.\n\nA district that runs Classroom, Chromebooks, Workspace identity and now Gemini has an integration surface with one vendor that would take three years and a bond issue to unwind. Students spend thirteen years learning to think inside one company's interface metaphors. Teachers build lesson libraries that live in one company's document format. The identity layer, the device layer, the learning management layer and the reasoning layer converge on a single supplier, and the exit cost climbs every term.\n\nThe strategic logic was described precisely by Chris Agnew, director of the Generative AI for Education Hub at Stanford University, speaking to the Christian Science Monitor about the vendors flooding into schools: “They want to capture the market. We've seen this many...” Kris Hagel, chief information officer at Peninsula School District in Washington, offered the plainer version of the same observation: “There has been an absolute gold rush.”\n\nThis is why the distinction between Google and Anthropic matters and must not be flattened.\n\nAnthropic supplies a tool to a teacher. The teacher verifies with a school email and identification, obtains a licence, and decides what to do with it. Whether and how educational data may be used is determined, by Anthropic's own statement, by district and state policies. The teacher is a human decision point with professional judgement, and crucially the teacher is an adult who consented on their own behalf.\n\nGoogle owns the platform on which the child's education physically runs. It owns the device, the account, the assignment stream, the file store and now the assistant. When Google changes a default, no human decision point stands between that change and the child. The administrator's earlier decision was retroactively repurposed.\n\nThat is not a moral verdict on either company. It is a structural fact about where accountability can be located. You can name the person who chose Claude. You cannot name the person who chose Gemini for a specific eleven-year-old, because no such person exists.\n\n## What Happens When Somebody Actually Asks the Students\n\nThe most useful recent research on this question comes from Germany, and its findings are quietly devastating for the assumptions embedded in these deployments.\n\nIn a paper titled “Mind the Trust Gap”, published to arXiv on 1 July 2026 and appearing in the Proceedings of the ACM on Human-Computer Interaction in October, Tomohiro Nagashima of Saarland University, Lisa Siegrist of the University of St Gallen, Niklas Scholz of RWTH Aachen, Shintaro Sato of Saarland, Martina Vincoli of ETH Zürich and Man Su of the Leibniz-Institut für Wissensmedien ran a speed-dating study using storyboards with sixteen school students and fifteen school teachers. It is a small qualitative study in one country, and it should be read as such. It is also, on its own account, the first attempt to systematically pair-match K-12 teacher and student views on AI decision-making control.\n\nThe headline misalignment is the trust inversion. Students trusted their teachers more than they trusted AI. Teachers believed the opposite, assuming students would trust AI more because they perceived its assessments as fairer than human judgement. The authors spell out the consequence: if teachers wrongly believe students prefer the machine, they will scaffold and intervene less often than students actually want.\n\nThe data-sharing findings are sharper still. Students expressed a strong desire to keep their own data private and to share only when necessary. They were willing to let their own teachers see it, and only their teachers, because they could see the pedagogical benefit. One thirteen-year-old participant put it directly: if you leave everything to yourself then nobody can help you, but if you share the biggest part with the teacher, the teacher can also help you.\n\nTeachers wanted broad access, and were reluctant to grant students control over sharing precisely because student choice would degrade the system's usefulness. One teacher with fifteen years of experience explained that the low-performing students in particular would not want to share their data, which would leave the teacher without an overview. Another, with two and a half years in post, suggested a student who does not know the teacher will be notified may be able to work without a certain amount of pressure.\n\nSit with that last one. It is an argument for concealing monitoring from a child in order to preserve the child's relaxed performance. It was offered in good faith, by a teacher trying to help.\n\nThe authors identify asymmetric power as a root cause. Teachers are responsible for outcomes, parents and curriculum, which pushes them towards monitoring and data access. Students are the subjects of that monitoring with limited agency. The same system reads as a support tool from one chair and a surveillance tool from the other. And the researchers found that students feared constant AI monitoring and data sharing as an additional pressure in their school lives.\n\nNow map that onto the American position. The Center for Democracy and Technology's October 2025 report, based on nationally representative surveys of 806 teachers, 1,018 parents and 1,030 students, found that thirty-one per cent of students had conducted back-and-forth conversations with AI for personal, non-schoolwork reasons on a device, tool or software provided by their school. Over two thirds of parents and students agreed that parents have no idea how students are interacting with AI. Thirty-eight per cent of students agreed it is easier to talk to AI than to their parents. Around one in five teachers and students had received any instruction on AI risks such as bias, misinformation or overreliance. Half of students agreed that AI use in class made them feel less connected to their teacher.\n\nChildren are already having personal conversations with school-issued AI. Their parents do not know. Almost nobody has been taught the risks. And in August the default moved down five year groups.\n\n## The Legal Architecture Was Built for Filing Cabinets\n\nFERPA is fifty-two years old, and it shows.\n\nThe school official exception is the load-bearing wall of every ed tech deployment in America, and it was designed for a world of contracted photocopying bureaux and student information systems. Its four conditions assume a vendor performing a discrete institutional function under direct institutional control. Applying “direct control with respect to the use and maintenance of education records” to a general-purpose reasoning model whose weights, training corpus, safety filters and system prompts are all outside the district's inspection is an exercise in legal metaphor rather than legal analysis.\n\nResearchers have caught the gap. In “Understanding How University Guidelines Address Privacy and Security Issues of Generative AI in Academic Settings”, Bei Yi Ng, Jiarui Li and Xinyuan Tong of the University of Edinburgh, Kevin Ye, Gauthami Yenne and Varun Chandrasekaran of the University of Illinois Urbana-Champaign, and Jingjie Li of Edinburgh analysed guidelines from forty-three universities across twelve countries. The paper is aimed at higher education, not K-12, and that limitation should be stated up front. Its diagnosis nonetheless transfers.\n\nThe authors observe that universities referencing FERPA for student data protection may leave leeway allowing third-party organisations such as generative AI vendors to use student metadata, a scenario disallowed under other regimes such as Canada's FIPPA, where personal data cannot be shared with third parties unless mandated by law. They describe generative AI adoption as occurring in a lemon market, where users lack visibility into model quality, risk and trade-offs. They note that in K-12 classrooms these risks are exacerbated, that children under thirteen carry additional protections under COPPA and FERPA, and that learning management systems nonetheless treat those children identically to everyone else. And they open the paper with the collapse of AllHere Education, the company behind the Los Angeles Unified chatbot, as the exemplar of what happens to sensitive student data when a vendor fails.\n\nThen there is the enforcement problem, which is the part almost nobody outside education law knows.\n\nFERPA has no private right of action. In Gonzaga University v. Doe, decided on 20 June 2002, the Supreme Court held that FERPA's non-disclosure provisions create no personal rights enforceable under section 1983. A parent whose child's records are mishandled cannot sue. The remedy is a complaint to the Department of Education's Student Privacy Policy Office, whose ultimate sanction is the withdrawal of federal funding, a nuclear option that has never once been deployed against a school district in the statute's history.\n\nCOPPA was supposed to backstop this for the under-thirteens, and here the news is genuinely bad. The Federal Trade Commission published its first substantial COPPA Rule amendments since 2013 on 22 April 2025, effective 23 June 2025 with full compliance required by 22 April 2026. The amendments expanded the definition of personal information, required separate parental consent for third-party disclosures, and tightened security requirements. What the Commission declined to do was codify its long-standing guidance permitting schools to authorise data collection on behalf of parents for educational technology. It left the mechanism as guidance, citing concern about conflicting with FERPA updates the Department of Education was expected to make.\n\nThe Department's own regulatory agenda listed that FERPA rulemaking, amending 34 CFR part 99 and covering disclosures to third parties such as commercial vendors, for September 2026. By mid-November no proposal had appeared in the Federal Register.\n\nSo the single legal device by which an American school substitutes its own consent for a parent's, for children under thirteen, in the year that two frontier laboratories moved into K-12 at scale, remains uncodified guidance that one agency declined to write down while waiting for another agency, which has now missed its own date.\n\nThe contrast with the United Kingdom is instructive without being triumphalist. The Information Commissioner's Office's Age Appropriate Design Code sets fifteen standards including privacy by default, data minimisation and no nudging, and requires services to differentiate protections across age bands rather than treating a six-year-old and a seventeen-year-old identically. The code does not apply to schools processing data for educational purposes. It may, however, apply to the ed tech provider, where the service meets the definition of an information society service. A regulator asking whether a default setting nudges a child towards lower privacy is asking exactly the question that went unasked on 10 August.\n\n## This Exact Film Has Screened Four Times Already\n\nAnyone treating August 2026 as unprecedented has a short memory. There are at least four prior reels.\n\nIn 2013, inBloom launched with roughly one hundred million dollars from the Gates Foundation and the Carnegie Corporation, offering to consolidate student records so teachers could track progress and personalise instruction. It ingested data across some four hundred fields, from test scores to special education enrolment to free lunch status. There was never a documented misuse. There did not need to be. Parents organised, states withdrew, and on 21 April 2014 inBloom announced it was shutting down. The wreckage produced more than four hundred student privacy bills in state legislatures, including the SOPIPA and SOPPA statutes still doing the work today.\n\nIn December 2015, the Electronic Frontier Foundation filed a complaint with the FTC alleging that Google's Chrome Sync feature, enabled by default on Chromebooks sold to schools, allowed the company to track the websites students visited, their search habits and saved passwords, in violation of the Student Privacy Pledge Google had signed.\n\nEnabled by default. On school devices. In 2015.\n\nIn February 2020, New Mexico Attorney General Hector Balderas sued Google, alleging that its G Suite for Education deployment collected children's physical locations, browsing and search histories, YouTube viewing habits, contact lists, passwords and voice recordings in violation of COPPA and state law. In September that year, District Judge Nancy Freudenthal dismissed the claims while granting leave to amend.\n\nAnd in 2023, Los Angeles Unified signed a six million dollar contract with AllHere Education for a chatbot called Ed. AllHere furloughed most of its staff and filed for Chapter 7 bankruptcy in July 2024. Chris Whiteley, the company's former senior director of software engineering, told officials that the application had included students' personally identifiable information in every chatbot prompt, including prompts where the data was irrelevant, that those prompts had been shared unnecessarily with third-party companies, and that they were processed on offshore servers. Founder Joanna Smith-Griffin was arrested on 19 November 2024 on charges including securities fraud, wire fraud and aggravated identity theft.\n\nFour episodes, three of them involving defaults or data flows nobody at the school had inspected. The pattern is not that ed tech companies are uniquely wicked. The pattern is that the consent architecture in American schools cannot detect a change until after it has happened, and cannot remedy one at all.\n\n## Nobody Asked the People Whose Data It Is\n\nThe final piece of research here is the one that reframes the whole argument, and it needs handling with care because it is easy to overclaim.\n\n“Participatory, not Punitive”, by Kaoru Seki, Manisha Vijay and Yasmine Kotturi, presented at the CHI conference in Barcelona in April 2026, is a small study. Three workshops in a graduate design course at the University of Maryland, Baltimore County, a minority-serving institution, with two student leaders facilitating and no faculty in the room. Eight participants. Ten policy recommendations. A zine.\n\nIt is not a national survey of K-12 policy. It is university level, tiny, and qualitative. Anyone citing it as proof that “most educational AI policies ignore students” is citing the authors' framing of the literature rather than their evidence.\n\nBut the framing is the interesting part, and it holds. Most institutional AI policies are written without the students they govern and organised around penalising misuse. The predominant enforcement instrument, AI detection, infers authorship probabilistically rather than comparing against known sources, which makes accusations more uncertain and more contestable than plagiarism findings ever were. Policies arrive without articulated pedagogical rationale, which teaches students to read them as arbitrary exercises of authority.\n\nAnd the finding the students surfaced that the administrators had not is the one that lands hardest: the double standard whereby students must disclose or abstain from AI use whilst faculty face no equivalent expectation. Instructors may use AI for teaching or grading without disclosure. Students may not use it without declaring it.\n\nNow hold that beside August 2026. A student who uses an undisclosed AI to draft an essay commits academic misconduct. A platform that enables an undisclosed AI inside that student's coursework, without asking the student, ships a feature.\n\nThe asymmetry is total, and it is the asymmetry that Nagashima and colleagues found students perceiving from the other direction in Germany, and that the Center for Democracy and Technology measured as a collapse in felt connection to teachers.\n\nAmerican students have noticed. Between 17 and 19 July 2026, at America's Youth AI Festival held on the UMass Boston and MIT campuses, high school students representing all fifty states convened a National AI Student Senate at the Edward M. Kennedy Institute and passed, by eighty-two votes to sixteen, a framework called the STUDENTS FIRST Act. Its provisions include the right to decline AI use and receive an alternative assignment, a requirement that students be informed about what data is collected and how it is used, a prohibition on entering identifiable or sensitive student information into AI tools, and a right to appeal findings of improper AI use with human review by multiple officials before discipline. The organising partners included Day of AI, MIT RAISE, AASA and the Kennedy Institute.\n\nTeenagers, given three days and a legislative chamber, produced a consent architecture more coherent than the one currently governing them.\n\n## The Question the Eleven-Year-Old Cannot Formulate\n\nReturn to the child at the centre of this, because the abstraction has a body.\n\nShe is eleven. She opens Classroom, as she has every school day for two years, and there is a tab that was not there in June. She does not know that a state board signed an agreement in a room in Salt Lake City, that a district opted in, that an administrator left a toggle where it was, or that a company in Mountain View changed what that toggle covers and has since announced it will change it again. She knows the school gave her a laptop and the laptop has a thing on it that answers questions. Schools do not, in her experience, put things on the laptop that she is not supposed to use.\n\nShe types a question about her homework. Then, because the research says roughly a third of students do, she types something else. Something about a friendship, or a fear, or a body. The Center for Democracy and Technology found that thirty-eight per cent of students find it easier to talk to AI than to their parents, and that over two thirds of parents have no idea any of this is happening.\n\nAsk the accountability question about that conversation and watch it dissolve. Google will point to the administrator console and the contractual commitments, both of which are real. The district will point to the state framework and local discretion, both of which are real. The state will point to local control, which is real. The federal statute offers her family no cause of action at all. The Federal Trade Commission has declined to codify the rule under which her school consented on her parents' behalf. The Department of Education can, in theory, defund her school district, which it will not do, because it never has.\n\nEvery actor behaved lawfully. Every commitment was honoured. And there is no one to ask.\n\nThat is the actual finding, and it is more uncomfortable than a villain would be. The problem is not that Google is harvesting children's homework to train models, because the evidence is that it contractually is not. The problem is that the American consent architecture for student data was built on the premise that an institution can substitute its judgement for a parent's on the assumption that the institution controls the technology. That premise held for filing cabinets. It held, barely, for student information systems. It does not hold for a general-purpose reasoning model updated on a vendor's release schedule, whose capabilities in November are not the capabilities the district evaluated in January.\n\nThe fix is not complicated to describe, and Google has now described part of it. Defaults involving minors should be off, requiring affirmative institutional action to enable. Once the November changes roll out, that will be Google's own position for its core student AI products, and it deserves credit as a real concession. What is still missing is everything that would make it durable. A fresh institutional action should be required whenever the eligible age band or the scope of a setting changes. Material changes to what a setting covers should trigger notification to parents, not merely to administrators; neither the August move nor the November one did. Basic tools such as search should not be bundled with AI, so that declining one costs a school the other. Districts should publish, as Illinois already requires, the list of vendors with access to student data. The FTC should codify the school authorisation exception, and the Department of Education should publish the FERPA rule it scheduled for September, rather than leaving the consent mechanism for the youngest children as guidance in a footnote. And students, who have now drafted their own framework and found it broadly reasonable, should be in the room. A default a company chose voluntarily is a default the company can unchoose.\n\nThe alternative is the pattern already established across four episodes and twelve years. A default gets set. Nobody notices. Something goes wrong, or nearly does. Parents organise. A company withdraws or a court dismisses. A legislature passes something. This autumn the cycle simply ran faster. The August default arrived on a Monday, in a product bulletin, under a subheading called “Rollout pace”. Its reversal arrived eighty-five days later, on a Tuesday, in a blog post, with the search box tied to it. The default for millions of children moved twice in one term, in opposite directions, at one company's discretion. Neither move went through a parent.\n\n## Sources and References\n\n1. Google Workspace Updates, “Gemini in Google Classroom is expanding to users of all ages, with contextualized Gemini starter prompts for students,” Google Workspace Updates Blog, 10 August 2026. [https://workspaceupdates.googleblog.com/2026/08/gemini-in-google-classroom-is-expanding-to-users-of-all-ages-with-contextualized-Gemini-starter-prompts-for-students.html](https://workspaceupdates.googleblog.com/2026/08/gemini-in-google-classroom-is-expanding-to-users-of-all-ages-with-contextualized-Gemini-starter-prompts-for-students.html)\n2. Kahekashan, “Google enables Gemini AI for students using its Classroom App,” The Hans India (syndicating New York Times reporting of 14 August 2026), 15 August 2026. [https://www.thehansindia.com/tech/google-enables-gemini-ai-for-students-using-its-classroom-app-1109903](https://www.thehansindia.com/tech/google-enables-gemini-ai-for-students-using-its-classroom-app-1109903)\n3. Youth AI Safety Institute, “What to know about Google turning on Gemini for schools,” Common Sense Media, August 2026. [https://institute.commonsensemedia.org/what-to-know-about-google-turning-on-gemini-for-schools](https://institute.commonsensemedia.org/what-to-know-about-google-turning-on-gemini-for-schools)\n4. Google for Education, “Gemini for Education.” [https://edu.google.com/intl/ALL_us/ai/gemini-for-education/](https://edu.google.com/intl/ALL_us/ai/gemini-for-education/)\n5. Anthropic, “Introducing Claude for Teachers,” anthropic.com, 14 July 2026. [https://www.anthropic.com/news/claude-for-teachers](https://www.anthropic.com/news/claude-for-teachers)\n6. Utah State Board of Education, “Utah State Board of Education Adopts Gemini for Education to Enhance Student Learning and Teacher Support,” 2 June 2026 (updated 3 June 2026). [https://www.schools.utah.gov/publicrelations/_publicrelations_/_2026/USBEAdoptsGeminiEducationPressRelease.pdf](https://www.schools.utah.gov/publicrelations/_publicrelations_/_2026/USBEAdoptsGeminiEducationPressRelease.pdf)\n7. Christian Science Monitor, “Education's AI 'gold rush' comes with risks for students,” csmonitor.com, 7 August 2026. [https://www.csmonitor.com/USA/Education/2026/0807/ai-risks-schools-students](https://www.csmonitor.com/USA/Education/2026/0807/ai-risks-schools-students)\n8. Tomohiro Nagashima, Lisa Siegrist, Niklas Scholz, Shintaro Sato, Martina Vincoli and Man Su, “Mind the Trust Gap: Identifying (Mis)alignments in Teacher-Student Views Toward Control and Agency in K-12 Classroom AI,” arXiv:2607.01506, 1 July 2026; Proc. ACM Hum.-Comput. Interact. 10, 6, Article CSCW124, October 2026. [https://arxiv.org/abs/2607.01506](https://arxiv.org/abs/2607.01506)\n9. Bei Yi Ng, Jiarui Li, Xinyuan Tong, Kevin Ye, Gauthami Yenne, Varun Chandrasekaran and Jingjie Li, “Understanding How University Guidelines Address Privacy and Security Issues of Generative AI in Academic Settings,” arXiv:2506.20463, 25 June 2025 (v2, 19 July 2026). [https://arxiv.org/abs/2506.20463](https://arxiv.org/abs/2506.20463)\n10. Kaoru Seki, Manisha Vijay and Yasmine Kotturi, “Participatory, not Punitive: Student-Driven AI Policy Recommendations in a Design Classroom,” arXiv:2604.10851, 12 April 2026; CHI '26, Barcelona, April 2026. [https://arxiv.org/abs/2604.10851](https://arxiv.org/abs/2604.10851)\n11. US Department of Education, “FERPA Frequently Asked Questions,” Student Privacy Policy Office. [https://studentprivacy.ed.gov/frequently-asked-questions](https://studentprivacy.ed.gov/frequently-asked-questions)\n12. Supreme Court of the United States, “Gonzaga University v. Doe, 536 U.S. 273,” 20 June 2002, via Legal Information Institute, Cornell Law School. [https://www.law.cornell.edu/supremecourt/text/01-679](https://www.law.cornell.edu/supremecourt/text/01-679)\n13. Federal Trade Commission, “Children's Online Privacy Protection Rule, Final Rule,” Federal Register, 22 April 2025. [https://www.federalregister.gov/documents/2025/04/22/2025-05904/childrens-online-privacy-protection-rule](https://www.federalregister.gov/documents/2025/04/22/2025-05904/childrens-online-privacy-protection-rule)\n14. Electronic Frontier Foundation, “Google Deceptively Tracks Students' Internet Browsing, EFF Says in Complaint to Federal Trade Commission,” eff.org, 1 December 2015. [https://www.eff.org/press/releases/google-deceptively-tracks-students-internet-browsing-eff-says-complaint-federal-trade](https://www.eff.org/press/releases/google-deceptively-tracks-students-internet-browsing-eff-says-complaint-federal-trade)\n15. New Mexico Department of Justice, “Attorney General Balderas Sues Google for Illegally Collecting Personal Data of New Mexican School Children,” February 2020. [https://nmdoj.gov/press-release/attorney-general-balderas-sues-google-for-illegally-collecting-personal-data-of-new-mexican-school-children/](https://nmdoj.gov/press-release/attorney-general-balderas-sues-google-for-illegally-collecting-personal-data-of-new-mexican-school-children/)\n16. Wendy Davis, “Google Defeats New Mexico AG's Lawsuit Over Students' Privacy,” MediaPost, 30 September 2020. [https://www.mediapost.com/publications/article/356337/google-defeats-new-mexico-ags-lawsuit-over-studen.html](https://www.mediapost.com/publications/article/356337/google-defeats-new-mexico-ags-lawsuit-over-studen.html)\n17. Valerie Strauss, “$100 million Gates-funded student data project ends in failure,” The Washington Post, 21 April 2014. [https://www.washingtonpost.com/news/answer-sheet/wp/2014/04/21/100-million-gates-funded-student-data-project-ends-in-failure/](https://www.washingtonpost.com/news/answer-sheet/wp/2014/04/21/100-million-gates-funded-student-data-project-ends-in-failure/)\n18. Parent Coalition for Student Privacy, “State Student Privacy Laws,” studentprivacymatters.org. [https://studentprivacymatters.org/state-legislation/](https://studentprivacymatters.org/state-legislation/)\n19. Information Commissioner's Office, “Introduction to the Children's code,” ico.org.uk. [https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/childrens-information/childrens-code-guidance-and-resources/introduction-to-the-childrens-code/](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/childrens-information/childrens-code-guidance-and-resources/introduction-to-the-childrens-code/)\n20. Center for Democracy and Technology, “Hand in Hand: Schools' Embrace of AI Connected to Increased Risks to Students,” cdt.org, October 2025. [https://cdt.org/wp-content/uploads/2025/10/FINAL-CDT-2025-Hand-in-Hand-Polling-100225-accessible.pdf](https://cdt.org/wp-content/uploads/2025/10/FINAL-CDT-2025-Hand-in-Hand-Polling-100225-accessible.pdf)\n21. Gallup, “Three in 10 Teachers Use AI Weekly, Saving Six Weeks a Year,” news.gallup.com, 2025. [https://news.gallup.com/poll/691967/three-teachers-weekly-saving-six-weeks-year.aspx](https://news.gallup.com/poll/691967/three-teachers-weekly-saving-six-weeks-year.aspx)\n22. Mark Keierleber, “Whistleblower: L.A. Schools' Chatbot Misused Student Data as Tech Co. Crumbled,” The 74, 2024. [https://www.the74million.org/article/whistleblower-l-a-schools-chatbot-misused-student-data-as-tech-co-crumbled/](https://www.the74million.org/article/whistleblower-l-a-schools-chatbot-misused-student-data-as-tech-co-crumbled/)\n23. AASA, The School Superintendents Association, “Students from All 50 States Produce National Framework for AI in America's Schools,” aasa.org, 3 August 2026. [https://www.aasa.org/news-media/news/2026/08/03/students-produce-national-framework-for-ai-in-america's-schools](https://www.aasa.org/news-media/news/2026/08/03/students-produce-national-framework-for-ai-in-america's-schools)\n24. Chris Phillips, “Designing responsible AI for learning,” Google, The Keyword, 3 November 2026. [https://blog.google/products-and-platforms/products/education/ai-learning-era-updates-2026/](https://blog.google/products-and-platforms/products/education/ai-learning-era-updates-2026/)\n25. Megan Cerullo, “Google will block search access for schools that ban AI use,” CBS News, 3 November 2026. [https://www.cbsnews.com/news/google-search-access-schools-ai-tools/](https://www.cbsnews.com/news/google-search-access-schools-ai-tools/)\n26. Jared Perlo and Tyler Kingkade, “Google to shut off search for schools that opt out of AI,” NBC News, 3 November 2026. [https://www.nbcnews.com/tech/tech-news/google-shut-search-schools-opt-ai-rcna607280](https://www.nbcnews.com/tech/tech-news/google-shut-search-schools-opt-ai-rcna607280)\n27. Anthony Izaguirre, “NYC, the nation's largest school system, bans AI for students through 8th grade,” Associated Press, 2 September 2026. [https://www.wsbtv.com/news/technology/nyc-nations-largest/TODW6DLLH4YRJJYMLRKDCODDQE/](https://www.wsbtv.com/news/technology/nyc-nations-largest/TODW6DLLH4YRJJYMLRKDCODDQE/)\n28. Google Workspace Admin Help, “History and Personalization in Search services for Workspace for Education users,” knowledge.workspace.google.com. [https://knowledge.workspace.google.com/admin/getting-started/editions/search-history-personalization-edu](https://knowledge.workspace.google.com/admin/getting-started/editions/search-history-personalization-edu)\n29. Office of Information and Regulatory Affairs, “Family Educational Rights and Privacy Act, RIN 1875-AA15,” Unified Agenda of Regulatory and Deregulatory Actions, Department of Education, Fall 2025. [https://www.reginfo.gov/public/do/eAgendaViewRule?pubId=202510&RIN=1875-AA15](https://www.reginfo.gov/public/do/eAgendaViewRule?pubId=202510&RIN=1875-AA15)\n\n**Tim Green**\n*UK-based Systems Theorist & Independent Technology Writer*\n\nTim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at [smarterarticles.co.uk](https://smarterarticles.co.uk), challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.\n\nHis writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.\n\n**ORCID:** [0009-0002-0156-9795](https://orcid.org/0009-0002-0156-9795)\n**Email:** [tim@smarterarticles.co.uk](mailto:tim@smarterarticles.co.uk)\n\nListen to the free weekly [SmarterArticles Podcast](https://www.smarterarticles.fm)", "url": "https://wpnews.pro/news/google-gemini-in-classrooms-who-consents-for-an-eleven-year-old", "canonical_source": "https://smarterarticles.co.uk/google-gemini-in-classrooms-who-consents-for-an-eleven-year-old?pk_campaign=rss-feed", "published_at": "2026-11-17 02:00:30+00:00", "updated_at": "2026-11-17 20:29:17.306064+00:00", "lang": "en", "topics": ["ai-policy", "generative-ai", "ai-products", "ai-ethics"], "entities": ["Google", "Gemini", "Google Classroom", "Chris Phillips", "Anthropic", "Claude for Teachers", "Derek Moore", "James O'Hagan"], "also_reported_by": [], "takeaway": null, "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/google-gemini-in-classrooms-who-consents-for-an-eleven-year-old", "markdown": "https://wpnews.pro/news/google-gemini-in-classrooms-who-consents-for-an-eleven-year-old.md", "text": "https://wpnews.pro/news/google-gemini-in-classrooms-who-consents-for-an-eleven-year-old.txt", "jsonld": "https://wpnews.pro/news/google-gemini-in-classrooms-who-consents-for-an-eleven-year-old.jsonld"}}