Two Senators Ask Google About the Missing Off Switch as Maryland's AI Policy Deadline Passes

Framing

Since Edition 32, this brief has sorted AI governance by instrument: a mayor's moratorium, a union and vendor standard, a state rule, an attorney general's court motion, eight state suits and a bankruptcy notice. This week the instrument is a letter, and the addressee is the vendor behind Classroom, Chromebooks, and Chrome. Edition 37 reported a Frederick County, Maryland, board member learning of a Gemini rollout after the fact and teachers who could not turn the feature off, and named the off switch as a procurement specification. On October 21, two senators from opposite parties asked Google, in writing, why that switch is sometimes missing.

Read the mechanism. A letter from two senators creates no obligation. It does something cheaper. It puts a question on the record with a date, November 2, and it lets every district compare its own settings against the allegation. The letter's strongest sentence, that Google's AI Overview and AI Mode chatbot "cannot be disabled and are the standard response to any search performed in the Chrome browser," is the senators' claim. Google's own help page, updated October 22, documents a control for Gemini in Classroom and says nothing about AI Mode or AI Overviews. That silence is the finding. A district cannot answer the senators' question about itself without reading its own administrator console. The Maryland story shows the same mechanism in reverse. Edition 35 called the compliance clock the new unit of governance. The clock ran out on October 22, and I found no statement of what happens to a district that misses it.

The research counterpoint is about what a design choice buys. Districts are told that a tutor built around self-regulated learning will outperform a general chatbot. In a preregistered randomized experiment with 144 high school students, learning improved in all three conditions, and the abstract reports no significant differences between conditions at follow-up. The scaffolded versions made students enjoy the work more and feel less bored. That is a real result, and it is not a learning result. A second paper tested 600 AI-generated multiple-choice questions against 15 item-writing guidelines and found that the tools often broke them, and that deeper prompts helped only slightly. The third reminds us what a polished metric can hide: a classroom camera system reports 93.46 percent emotion classification and no learning outcome. Edition 34 counted seven of sixteen widely used products with independent achievement studies. The pattern holds.

The timing argument is a thirty-day window. Between now and November 2, a district can map every AI feature on student accounts and managed devices, write the default and the control path next to each, and be ready when a board member or reporter reads the senators' letter. Before December, a district under a state mandate can publish a dated adoption calendar, which Anne Arundel did and its neighbors had not. Waiting costs you the chance to find a missing switch yourself. It also leaves you explaining a deadline you didn't meet to a parent who has read about another district that also missed it.

Top Research and Policy Signals

1. Two Senators Asked Google Why AI Is the Default and, in Some Cases, Cannot Be Disabled on Licensed Education Devices, and Google's Own Help Page Confirms a Control for One Product and Is Silent on the Others

Source type. Congressional letter requesting information, not law or rule, October 21, 2026.

Citation (APA). Blumenthal, R., & Hawley, J. (2026, October 21). [Letter to Sundar Pichai, Chief Executive Officer, Alphabet Inc. and Google LLC, regarding A.I. tools in classrooms]. U.S. Senate. blumenthal.senate.gov knowledge.workspace.google.com

On October 21, 2026, Senators Richard Blumenthal of Connecticut and Josh Hawley of Missouri signed a letter to Sundar Pichai. It requests answers to nine questions "by November 2nd, 2026." Question 2 reads: "Why has Google set A.I. product integration as the default and, in some cases, made it impossible to disable these features on licensed education devices?" The letter states that "Google's A.I. Overview and A.I. Mode chatbot cannot be disabled and are the standard response to any search performed in the Chrome browser."

The other questions ask for Google's research on harm to children's well-being, concerns districts raised and Google's response, what Google has spent promoting AI in schools, cases where classroom AI encouraged harmful behavior or emotional dependence or missed suicidal or self-harm ideation in minors, and steps against sycophantic or emotional chatbot conversations with children. The letter relies on outside reporting that it attributes to others, including a Wall Street Journal account of a Maryland student hospitalized after developing dependence on Google's AI, and a figure of more than 150 million Classroom students. I did not verify those underlying reports.

Google's help page says Gemini in Classroom is on by default and available to users of all ages based on the institution's age-based access settings. An administrator sets it under Generative AI, Gemini for Workspace, Feature access in Classroom, by organizational unit or group, and users under 18 whose access was previously turned off stay off. The page does not mention AI Mode or AI Overviews.

Study context. A letter is a request, not a rule, and I found no legal duty on Google to answer by November 2. The senators make the factual claims about Google, and the help page does not confirm or refute the statement about Search features on managed Chromebooks. I found no public Google response.

Leadership implication. Before November 2, have your chief technology officer deliver a one-page table to your cabinet listing every AI feature reachable from a student account or a managed Chromebook: the default state, the administrator control path, the organizational units where it is on, and whether a control exists at all. Where the answer is that it cannot be disabled, put that sentence in the board memo and add a required disclosure to your next data privacy agreement renewal, naming a Google contact who answers in writing.

2. Maryland's Local AI Policy Deadline Passed on October 22 With Several Baltimore-Area Districts Reporting No Adopted Policy, and the One National Count of State Guidance Says Only Maryland Requires Tools to Be Vetted for Outcomes

Source type. Two named news reports on a state statutory deadline, October 21 and 23, 2026, and an institutional report (Center on Reinventing Public Education, not peer-reviewed), October 21, 2026.

Citation (APA). Calefati, J. (2026, October 23). Anne Arundel parents, teachers urge school board to restrict AI in classrooms. The Banner. thebanner.com Paul, A. (2026, October 21). [Story on Anne Arundel County parents' and teachers' concerns about AI use in schools]. WJZ/CBS Baltimore. cbsnews.com Center on Reinventing Public Education. (2026). The state of the American student 2026 [Report and fast facts]. Arizona State University. stateofthestudent.crpe.org and crpe.org Maryland General Assembly. (2026). Senate Bill 720, Chapter 634, Artificial Intelligence Ready Schools Act. mgaleg.maryland.gov

Maryland's Artificial Intelligence Ready Schools Act, covered in Edition 35, requires each local school system to have an AI policy aligned with state guidance within 120 days of the Department's release of that guidance. The Department's framework is marked "Last updated: June 2026" and carries no release day. CBS Baltimore wrote that "The deadline for those policies is Oct. 22," and The Banner described a Thursday deadline set by state legislators. The date fits a late June release by arithmetic, which the Department has not confirmed.

The Banner reported on October 23 that Anne Arundel County Public Schools was among a handful of Baltimore-area school districts that missed the deadline, and that Baltimore City, Baltimore County, and Howard County "have not yet introduced AI policies, let alone passed them." Anne Arundel's board considered a first draft on October 21. CBS Baltimore reports a second reading on December 2 and planned adoption on December 16; The Banner says mid-December. The Banner also reports that Howard County gave high school students access to Gemini earlier this year.

The Center on Reinventing Public Education released its fifth annual State of the American Student on October 21, per The 74. Its fast facts count 36 of 50 states that published or endorsed K-12 AI guidance since 2023 and 14 that did not, and report that 1 of the 36, Maryland, requires AI tools to be vetted for academic outcomes. The report says most guidance is advisory, and almost no state has funded districts to meet it.

Study context. These are news reports, not board records or Department statements, and I could not open the Anne Arundel board page. The statute, as I read it, requires districts to use a state rubric to evaluate tools; I could not confirm an academic outcome test, so the vetting characterization is the Center's. I found no statement from the Department on consequences.

Leadership implication. If your state has set a date, publish your own adoption calendar now with the first reading, second reading, adoption date, and a named owner for each. Anne Arundel's sequence is a usable template. Ask your board attorney in writing what the state can do to a district that misses, because I could not find an answer. Add to the draft policy a current inventory of AI tools already live in the district, since Howard County's Gemini access shows a policy can arrive after the tool.

3. A Preregistered Trial With 144 High School Students Reports That Learning Improved in All Three Conditions and That the Scaffolded Chatbots Did Not Pull Ahead of ChatGPT at Follow-Up, Only on Enjoyment and Boredom

Source type. Peer-reviewed journal article, British Journal of Educational Technology, online October 18, 2026. Published version; an earlier preprint exists and differs (see below).

Citation (APA). Diordiev, E.-A., Stenberdt, V. A., & Makransky, G. (2026). Prompt-based emotional scaffolding in generative AI: Effects on self-regulated learning and socio-emotional outcomes. British Journal of Educational Technology. Advance online publication, article bjet.70094. doi.org

The authors, all at the Department of Psychology, University of Copenhagen, report a preregistered experiment in an authentic educational setting. A total of 144 high school students were randomly assigned to one of three conditions: standard ChatGPT, a chatbot built to scaffold self-regulated learning (SRLbot), or the same chatbot with emotional support added (Emotional SRLbot). Data were collected at pretest, posttest, and a four-week follow-up.

Per the abstract, conceptual knowledge and transfer improved in all conditions, with "no significant condition differences in post-to-follow-up change." Both scaffolded chatbots were associated with higher self-reported enjoyment and lower boredom than ChatGPT. The Emotional SRLbot produced more performance phase and self-reflection behavior in the trace data. The authors describe the results as preliminary and call for longitudinal research.

A 2025 preprint of this work, posted May 28, 2025 on PsychArchives under a different title and not peer-reviewed, said only the ChatGPT group declined significantly at follow-up and that only the emotional version beat ChatGPT on enjoyment and boredom. The published abstract says something different on both points. This brief cites the published wording. It reminds us that peer review can change what a study is allowed to claim.

Study context. I read the Crossref record, parts of the abstract, and the preprint page; the Wiley page and full text did not load, so I cannot report effect sizes, per-arm counts, task length, or the study's country. The abstract does not name a country, and the authors' affiliation does not indicate the study site. The study measured learning on researcher-designed tasks over four weeks.

Leadership implication. Put one question into your next tutoring or chatbot renewal and every vendor demo: compared with what? If the comparison is a general chatbot used the way students already use it, ask for the learning difference at follow-up, not only enjoyment and engagement. Require the vendor to state the design, the comparison condition, and the follow-up interval in writing before the pilot starts, and file that statement with the renewal so a board member can read it later.

4. Three AI Tools Wrote 600 Multiple Choice Questions and Frequently Broke Item-Writing Guidelines, So the Human Review Step Is the Control, and the Study Did Not Measure Outcomes for Students With Disabilities

Source type. Peer-reviewed journal article, Journal of Special Education Technology, online first October 21, 2026.

Citation (APA). Frey, B. B., Murray, A., & Mosher, M. A. (2026). The validity of AI-generated multiple-choice questions for students with disabilities. Journal of Special Education Technology. Advance online publication. doi.org

Frey, Murray, and Mosher, of the University of Kansas Department of Educational Psychology, asked ChatGPT, Copilot, and Gemini to generate 200 multiple-choice items each, under default and deeper prompt settings, for 600 items in total. They scored each item against 15 item-writing guidelines and compared the tools with chi-square tests and analysis of variance.

The most common violations were incomplete stems and non-randomized placement of the correct answer. ChatGPT had the highest overall error rate, and Copilot and Gemini the lowest, with statistically significant differences among the tools. Deeper prompting improved item quality only slightly. The authors conclude that "educators must review and revise them to ensure their quality."

The page does not provide the model versions used or the error percentages, which sit behind the paywall. The abstract page does not state the source of the 15 guidelines, although the reference list includes classic item-writing rules, and I do not assert the source.

Study context. This is a rule-compliance study of generated items. It has no field test with students, no difficulty or bias statistics, and no outcome measure, so it cannot say whether an item that follows the guidelines is fair to students with disabilities. I read the abstract and the Crossref record, not the full text.

Leadership implication. If teachers or a vendor in your district generate quiz or assessment items with AI, write the human review into the assessment procedure now: who reviews, against which checklist, and where the reviewed item is stored before it reaches a student. For students with IEPs or 504 plans, send the checklist to the special education director, and add a procurement clause that bars a tool from delivering AI-generated items directly to students without a review record.

5. A Peer-Reviewed Camera System Reports 93.46 Percent Emotion Classification and Turns Student Faces Into an Attentiveness Score Without a Learning Outcome or a Check Against Human Observers

Source type. Peer-reviewed journal article, Frontiers in Imaging, volume 5, published October 13, 2026. Two university classrooms in India; not K-12 evidence. Used as a counter-evidence exhibit.

Citation (APA). Davda, B., Vahora, S., Maltare, N., & Patel, H. (2026). Edge-AI-enabled real-time classroom attentiveness monitoring via facial emotion recognition on Raspberry Pi. Frontiers in Imaging, 5, Article 1972869. doi.org

The authors, at Government Engineering College Modasa in Gujarat, built a system that captures a classroom frame once per minute, detects faces with a pre-trained RetinaFace model, classifies seven emotions with DeepFace, and converts the counts into a weighted attentiveness score from 0 to 1, with thresholds labeling classes Low, Medium, or High. They collected their own data in two university classrooms over eight sessions of about 50 to 55 minutes, with 60 to 90 students per session.

They report face detection of 95.67 percent against 89.46 percent for a three-layer convolutional baseline, emotion classification of 93.46 percent against 78.42 percent, and 62 milliseconds per frame on a Raspberry Pi 5. The weights that turn emotions into attentiveness are set by hand.

What the paper does not contain matters more. It reports no learning outcome, no validation of the attentiveness score against human observers, and no limitations section. The paper attributes ethics approval to the head of the Information Technology department rather than an ethics committee, and it reports written consent. The paper states that it does not write facial images, embeddings, or templates to permanent storage. It also contains internal inconsistencies: the conclusion says 620 milliseconds per frame, and one section gives a 600-second sampling interval where others say 60.

Study context. University students in India are not K-12 students, and the paper tests a prototype's recognition accuracy, not whether a score improves teaching or learning.

Leadership implication. Adopt a standing rule for the technology review committee: no student-facing or classroom-facing system that infers attention, emotion, or engagement from faces or voices is piloted without a written validation against trained human observers, a learning outcome measure, and a privacy review of retention. Send the rule to the same vendor questionnaire you use for AI tools, and ask whether the product reports accuracy only, because accuracy of a label is not evidence that the label means anything.

Emerging Strategic Themes

Theme 1. When the switch is missing, the question moves to Washington. Edition 37 named the off switch a procurement specification after a Maryland board learned of a Gemini rollout late. This week two senators asked the vendor why the switch is sometimes absent. A district cannot answer for itself without reading its administrator console, so read it. Record each AI feature's default, its control path, and the contract sentence that guarantees a control, and bring that table to the board before a reporter does.

Theme 2. A deadline without a named enforcer is a calendar entry. Edition 35 called the compliance clock the unit of governance. Maryland's clock ran out on October 22; several Baltimore-area districts had no adopted policy, and I found no statement of consequences. A clock matters only if someone is empowered to act when it stops. Ask your state agency in writing who enforces your state's AI deadline, and publish your own dated calendar regardless of the answer.

Theme 3. The comparison group is the claim. Edition 34 counted seven of sixteen products with independent achievement studies, and Edition 36 warned that the tool is not the treatment. This week, a purpose-built tutor was compared with plain ChatGPT and showed no learning advantage at follow-up, only more enjoyment. A vendor claim of better is empty until it names better than what. Require the comparison condition, the outcome, and the follow-up interval in writing before any pilot starts.

Theme 4. A quality check is not a learning measure, and a human review step is a control. A camera system reports 93.46 percent emotion accuracy with no learning outcome, and AI-generated test items broke writing guidelines often enough that the authors tell teachers to review every one. Edition 38 warned that agreement among models is not validity. The shared lesson is that automated confidence is cheap and review is the only control with a record. Fund the review step and require it in the contract.

What Was Not Found

No public source confirms which Google AI features an administrator can and cannot disable on managed Chromebooks. I read the senators' letter, both press releases, and Google's Gemini in Classroom administrator page. The page documents a control for Gemini in Classroom and does not mention AI Mode or AI Overviews, and I found no Google response to the letter. Districts are signing and renewing Chromebook and Workspace agreements without a vendor-published list of which AI features can be turned off.

No source states what Maryland, or the Department, will do about a district that missed the October 22 deadline. I searched the Maryland State Department of Education pages and news reports and found no enforcement statement or confirmed guidance release date. The claim that Maryland vets tools for academic outcomes rests on the Center on Reinventing Public Education's count, whose method I could not read. Other states that are setting compliance dates are copying a model whose enforcement is unknown.

No new randomized K-12 trial of a generative AI tool with a standardized achievement outcome appeared in the window. The strongest design I found was the Diordiev trial: 144 high school students, three conditions, researcher-designed learning measures, a four-week follow-up, and no effect sizes in the abstract. I searched Consensus, journal indexes, arXiv listings, and the EdWorkingPapers list for October 12 to 25. Districts are renewing tools on evidence of this size. This persistent gap carried over from Editions 34, 37, and 38.

No K-12 study of attention, engagement, or emotion detection measured learning or validated its scores against human observers. The new detection papers I found, including the Davda study, report recognition accuracy in university or adult samples. I searched for classroom surveillance, engagement detection, and proctoring work and found none in K-12 that links the score to learning or reports a privacy governance framework. Vendors can sell these systems on accuracy alone.

No new empirical study of AI for English learners, and no student-outcome study for students with disabilities, appeared in the window. The only special education item was Frey and colleagues, which tested the quality of generated questions, not student outcomes. For English learners and elementary literacy, I found one scoping review whose date I could not clarify, and which draws mainly on Asian English-as-a-foreign-language settings. Districts apply AI accommodations to these populations on evidence not gathered from them.

No Georgia district AI board action in the window could be confirmed, and Lane 7 coverage was partial. Simbli listings showed Gwinnett work session and business meetings on October 20 and 22, Forsyth on October 20, and DeKalb on October 26, but agendas were not readable, and no AI item could be verified. The Atlanta Public Schools Policy Review Committee met October 21; its minutes were unpublished, and I could not confirm whether draft Policy IFBI moved. Cobb and Cherokee were not reached, and no Tier 2 district AI item surfaced.

The pattern is that deadlines and defaults are outrunning the facts a district needs to comply. A Senate letter asks about controls no public document lists, a state clock ran out with no stated consequence, and the evidence on whether scaffolded tools raise learning is a single short trial. The correct response is not delay. It is monitoring, documentation, and exit ramps: a feature control table for each vendor, a dated adoption calendar for each mandate, and a date by which each pilot either produces its own comparison data or ends.

Novo Executive Summary

On October 21, two senators asked Google why AI is the default and sometimes cannot be disabled on licensed education devices, and the next day Maryland's local policy deadline passed with several Baltimore-area districts reporting no adopted policy. The research record shows why neither question has an easy answer. A preregistered trial of 144 high school students found learning improved in every condition, and no follow-up advantage for scaffolded chatbots; a study of 600 AI-generated test questions found frequent errors that only a human reviewer would catch; and a camera system reports 93.46 percent emotion accuracy without measuring learning. The conclusion is not which vendor to trust. It is architecture: who holds decision rights over each AI feature and whether a control exists; what the procurement specification requires for comparison conditions and review steps; how evidence is defined before a pilot starts; and what training lets staff read an administrator console and a vendor claim. Novo Innovative Pathways works with district leaders to build that governance architecture, role-based AI literacy, and implementation strategy, so a district can answer a senator's question about itself before anyone asks it.

Watch This Week

• Monday, October 26, 2026, Colorado Attorney General. Comments close at 11:59 p.m. Mountain and a hearing is reported for 10 a.m. on the interim draft rules under SB 26-189 and HB 26-1263, effective January 1, 2027. Carried forward from Edition 38; I could not confirm whether the rules reach public K-12 districts.
• Thursday, October 29, 2026, Florida. Amended Rule 6A-1.0957, Internet Safety Policy, takes effect, with the district deadline of July 1, 2027 reported in Edition 34. Carried forward from Editions 34 through 38; I did not re-read the rule this week.
• Friday, October 30, 2026, 10:00 a.m. Eastern, U.S. Bankruptcy Court, District of Delaware, In re Character Technologies, Inc., Case No. 26-11603-BLS. A docket tracker lists the bidding procedures hearing for that time and reports the states' October 23 objection as Docket No. 66, with Florida's joinder at Docket No. 58; a news report counts the objecting states inconsistently (31 and 32 plus the District). These are rung three and not checked against the docket. Proposed dates reported by PPC Land: bid deadline November 20, auction November 23, sale hearing November 24.
• Monday, November 2, 2026, Google. The date by which Senators Blumenthal and Hawley requested written answers to nine questions. I found no public response and no sign Google must publish one. Also November 2: the Atlanta Public Schools board meeting, where draft Policy IFBI may appear; I could not confirm the Policy Review Committee's October 21 outcome. Carried forward from Editions 32 through 38.
• Wednesday, December 2 and Wednesday, December 16, 2026, Anne Arundel County Public Schools. CBS Baltimore reports a second reading of the draft AI policy on December 2 and planned adoption on December 16; The Banner says mid-December. Watch whether Baltimore City, Baltimore County, and Howard County introduce policies, and whether the State Department states any consequence for the missed October 22 deadline.
• Friday, November 13, 2026, Los Angeles Unified and Monday, November 9, 2026, Kansas State Board. The Generative AI Ad Hoc Committee was postponed to November 13 per Edition 38. The Kansas board received its instructional technology item on October 13 with no vote, and the packet lists November 9 as the next meeting. Utah's HB 273 requires its model policy before December 1. Carried forward from Editions 36 through 38.
• Date not set, Collin County, Texas. Texas Attorney General Ken Paxton sued OpenAI and Sam Altman on Friday, October 23, in Collin County district court, separately from the eight state suits in Edition 38. The Texas Tribune reports relief that includes blocking minors' access and a court-appointed evaluator for new models; it names no case number, statutes, or schools, and I could not open the Attorney General's release or the petition.

Sources

Governance and Policy

Blumenthal, R., & Hawley, J. (2026, October 21). [Letter to Sundar Pichai, Chief Executive Officer, Alphabet Inc. and Google LLC, regarding A.I. tools in classrooms]. U.S. Senate. blumenthal.senate.gov

Blumenthal, R. (2026, October 21). Blumenthal & Hawley demand answers from Google about aggressive AI push in classrooms [Press release]. blumenthal.senate.gov

Hawley, J. (2026, October 21). Hawley, Blumenthal demand answers from Google about rampant A.I. push in classrooms [Press release]. hawley.senate.gov

Google. (2026). Manage access to Gemini in Classroom [Administrator help page, last updated 2026-10-22 UTC]. Google Workspace Admin Help. knowledge.workspace.google.com

Maryland General Assembly. (2026). Senate Bill 720, Chapter 634. mgaleg.maryland.gov

Calefati, J. (2026, October 23). Anne Arundel parents, teachers urge school board to restrict AI in classrooms. The Banner. thebanner.com

Paul, A. (2026, October 21). [Story on parents' and teachers' concerns about AI use in Anne Arundel County schools]. WJZ/CBS Baltimore. cbsnews.com

Klibanoff, E. (2026, October 23). Texas Attorney General Ken Paxton sues ChatGPT-maker OpenAI. The Texas Tribune. texastribune.org

In re Character Technologies, Inc., Case No. 26-11603-BLS (Bankr. D. Del.). Docket summary and hearing notice, Dkt. 50, 58, and 66. elevenflo.com Rijo, L. (2026, October 23). Thirty-two states say Character.AI sale rules block bids that would beat Disney. PPC Land. ppc.land

Florida Department of State. (2026). Rule 6A-1.0957, Internet Safety Policy. flrules.org

Colorado Department of Law. (2026). Automated decision-making technology act and chatbot safety act rulemaking. coag.gov

Simbli board listings, Georgia districts (Atlanta Public Schools S=36031014; Gwinnett S=36031119; Forsyth S=4069; DeKalb S=4054). simbli.eboardsolutions.com

Research, Peer-Reviewed

Diordiev, E.-A., Stenberdt, V. A., & Makransky, G. (2026). Prompt-based emotional scaffolding in generative AI: Effects on self-regulated learning and socio-emotional outcomes. British Journal of Educational Technology. Advance online publication, article bjet.70094. doi.org

Frey, B. B., Murray, A., & Mosher, M. A. (2026). The validity of AI-generated multiple-choice questions for students with disabilities. Journal of Special Education Technology. Advance online publication. doi.org

Davda, B., Vahora, S., Maltare, N., & Patel, H. (2026). Edge-AI enabled real-time classroom attentiveness monitoring via facial emotion recognition on Raspberry Pi. Frontiers in Imaging, 5, Article 1972869. doi.org

Research, Preprint (Not Peer-Reviewed)

None cited as evidence this week. The 2025 PsychArchives preprint of the Diordiev study is described in Signal 3 only to note a discrepancy; it is not peer-reviewed and is not relied on. No other preprint met the window, K-12, and verification tests.

Institutional Report (Not Peer-Reviewed)

Center on Reinventing Public Education. (2026). The state of the American student 2026. Arizona State University. stateofthestudent.crpe.org and crpe.org Institutional report, not peer-reviewed.

Napolitano, J. (2026, October 21). [Coverage of the Center on Reinventing Public Education's State of the American Student 2026]. The 74. the74million.org