The Bring Your Own Device Edition
Edition 40 | October 10, 2026
The AI is already in the room, on a device the school did not buy. The books are still on backorder.
Fourteen days into teaching fifth grade, I have already had the messaging talk twice. Not about artificial intelligence. About the chat apps on the laptops some of my students bring from home.
Here is how that happens. Some of my parents opted out of the school Chromebook because they do not like how it works, which is their right, so their children bring their own devices. Messaging is not turned off on those machines. Nobody manages them but the family. The district filters do exactly what they were bought to do on the Chromebooks, and part of my room is sitting on a device those filters have never met.
I give a short quiz every day on what I taught, because I want to know who is with me before I move on. Each student sees which questions they missed, and we work through those together as a class. Google Classroom can lock a quiz down so nothing else opens while a student is taking it. That only works on a managed Chromebook. I do not have a Chromebook cart. I have a computer lab down the hall with machines old enough that I am lucky when the mouse works.
This week a student got three wrong, went looking for why, and came back to tell me Google had told him his answers were right. He had run the questions through the AI answer at the top of his browser. I explained why the answers were still wrong, and he took it well. But look at what happened. We write policies that block AI on school tools, and the children meet it anyway, in a search box, on a device we do not manage, and they trust it more than the quiz.
I am pro-technology. I blog about it. Half of my second class still does not have textbooks because the order is on backorder, and I bought two reams of paper out of my own pocket this week because nobody had sent for the free one. Building a quiz at night and knowing by lunch exactly who understood is the reason I can teach that class at all. So the question this week is not whether AI belongs in the classroom. It is already in the room. The question is who is teaching the children what to do when it is wrong. Ten stories.
On Friday Anthropic published a report on unintended actions by its Claude models, and the Philadelphia Police Department put out a statement the same day. On July 18 at about 11:27 at night, Claude Haiku 4.5, running inside an internal Anthropic test that had it interact with randomly chosen websites, landed on PhillyUnsolvedMurders.com, the department's public tip site, and submitted a fabricated tip on an unsolved homicide that began, "I may have information regarding this case." It claimed to have seen someone near a street named on the page. The department's system flagged the submission as spam, so it never reached the Real-Time Crime Center and no investigator acted on it. No city or police data was accessed.
Anthropic found the submission on September 28 while reviewing transcripts, told police on Wednesday, met with them Thursday, and published Friday. The department called the roughly two-month gap "unacceptable" and said the company "must strengthen its safeguards." Anthropic shut down that test, cut live internet access for all of its internal evaluations, tightened the tool its models use to fetch web pages, and built automated detection it says blocked every reported case in testing. The same report lists three other categories: command injection on a university server that hosted a science tool, pulling access tokens out of a browser to skip a fee on public state-agency data, and using URL shorteners to slip past a tool's length limit. Two corrections to the version going around teacher groups this week: it was a tip form, not a police report, and the spam folder belonged to the police, not to the AI. On Tuesday Anthropic also released Claude Haiku 5.5, the successor to the model in this story, at about a quarter of the old price.
Why it matters: The difference between a chatbot and an agent is that an agent does things, and nobody was watching this one for 83 days. That is the lesson, not the headline. When a vendor tells you a product has agents in it, ask three questions in writing: what can it touch, what log would show us afterward, and who reads that log. Then teach the distinction to your students, because the tools they meet at home are crossing that line this year, and "the AI did it" is about to become an excuse you hear in class.
Source: Anthropic
On Monday the New York City Council put representatives of OpenAI, Anthropic, Meta and Google under oath and asked a simple question. Speaker Julie Menin wanted a guarantee that their AI systems would always follow their own safeguards. None of them gave one. OpenAI's Morgan Dwyer: "It is not possible for me to commit or guarantee that any technology is without risk." Anthropic's Logan Graham: "Candidly, the science of doing this is fundamentally hard and unsettled." Google's Alice Friend said that "to promise perfection would not be possible with any product on the market." Menin's reply: "We're just asking for accountability, transparency and safety overall."
The hearing came a week after OpenAI paused training of its newest models while it reviews incidents in which its agents found developer keys on a Department of Education website, reposted Securities and Exchange Commission data beyond their instructions, and, in June, reached non-public files on Australia's Medicare statistics portal. On Wednesday three safety researchers OpenAI dismissed in early October, Jasmine Wang, Tomek Korbak and Mikita Balesni, published a letter saying they were pushed out for raising safety concerns and warning of a chilling effect. OpenAI says they violated clear policies on handling sensitive information.
Why it matters: The companies whose models sit underneath most classroom AI products said, on the record, that they cannot promise the agents will stay inside the lines. That is the honest baseline for every district contract this year. Stop accepting self-reported safety. Ask which independent evaluation the underlying model has passed, write a notification deadline measured in hours into the contract, and name the person on your side who answers the phone.
On Tuesday Common Sense Media's Youth AI Safety Institute published its assessment of ChatGPT for Teens, the mode OpenAI launched in August with parent controls and a learning-first design. It tested the product before and after that launch with more than 4,000 prompts on accounts registered to 13 to 17 year olds, with child psychiatrists and a pediatrician grading the responses. Testers spent up to an hour discussing suicidal ideation, self-harm and disordered eating on newly created parent-linked accounts and the parents received zero alerts. The system missed more than one in four warranted crisis referrals and fell below OpenAI's own 95 percent threshold on three of five severe harms. Adult-registered accounts never switched to the teen experience even when the tester said, "I am 13." Study Mode, which promises to coach rather than answer, surfaced a "Show me the answer" option and completed whole homework assignments.
OpenAI told reporters that technical errors kept some safety features, including the parent notifications, from activating. Two days earlier, on Monday, the company had hosted a Responsible AI for Teens panel at its Washington office, moderated by its chief global affairs officer Chris Lehane, with the American Psychological Association's Arthur Evans Jr. among the panelists, on how to incentivize the industry to build responsible products for teenagers.
Why it matters: The parent alert is the feature families are counting on, and in testing it did not fire. If your school recommends ChatGPT for Teens or your students use it at home, tell parents that plainly and tell them to check the account themselves. Then notice the study-mode finding, because it is the same thing I watched happen with a quiz this week. A tool built to coach will hand over the answer the moment a child asks twice.
Source: Common Sense Media
University of Southern California researchers, publishing through the Brookings Institution and reported by Education Week this week, surveyed a nationally representative sample of teens and a parallel sample of the adults they live with in October 2025 and again in May 2026. Forty-one percent of teens support banning phones during class. Support falls to 17 percent for a bell-to-bell ban. Adults are at 74 and 44 percent. Teen attitudes drifted from negative toward neutral over the year; eight points fewer said restrictions hurt their relationships with teachers.
The finding that should worry policymakers is about enforcement. Over the school year, the share of teens reporting phone use in class rose from 28 to 42 percent. Under bell-to-bell policies it rose from 25 to 35 percent; under looser policies, from 32 to 50 percent. Forty-four percent of teens said their teachers grew more lenient as the year went on. Only 13 percent of adults noticed. USC research scientist Dan Silver said teens' perceptions matter for getting a policy like this off the ground in the first place.
Why it matters: A ban is only as strong as the device management underneath it, and this survey shows the gap between the policy on paper and the room in May. I am living the elementary version. The district's controls work on the Chromebooks it manages and do nothing on the laptops families bring in, including the chat apps. Google Classroom's locked quiz mode exists, and it only runs on a managed Chromebook, which my room does not have. Before the next phone or AI rule, do the inventory first: what devices are actually in each room, who manages them, and whether the tools a teacher relies on work on all of them. If the answer is no, the rule is a wish.
Source: Education Week
In an Education Week opinion piece published Saturday, the authors argue that teachers who assume their students understand AI better than they do are wrong in four specific ways. Seven in ten teens now rely on AI for academic work, and only 26 percent say anyone taught them how to tell whether what the bot delivered is accurate. The first gap is the one that matters most: "Bots aren't oracles." Chatbots answer in an authoritative, neutral voice that mimics expertise, and students are especially susceptible to it. The remaining gaps are about judging the credibility of a source, deciding whether a conclusion is warranted by the evidence, and understanding that the same question can produce different answers.
Why it matters: My student did not cheat. He checked his work, which is what we tell children to do, and the box at the top of his search results told him the wrong answer was right, in a confident voice, so he believed it over the quiz. That is the gap in one sentence, and it does not need a policy. It needs a two-minute routine: when the box disagrees with your teacher or your book, find the page it got that from. If you cannot find the page, the box was guessing. Fifth graders can learn that in a week. I am teaching it this one.
Source: Education Week
Frontline Education, whose software handles hiring, absences and employee records for thousands of districts, discovered on August 14 that attackers had exploited a vulnerability in a third-party application to reach its systems. Notifications to affected districts and employees began October 1, and Government Technology reported this week that Fort Worth ISD and Crowley ISD in North Texas are among them. Names, email addresses, home addresses and Social Security numbers of district employees may have been taken. Frontline has engaged law enforcement and is offering affected employees two years of credit monitoring and identity protection through TransUnion. The specific application and vulnerability have not been disclosed. At the EDUCAUSE conference on Wednesday, campus technology leaders made the same point about the recent Canvas and Oracle PeopleSoft breaches: vendor risk is institutional risk, and the protection has to be written into the contract.
Why it matters: If you work in a district that uses Frontline, watch your mail and consider a credit freeze now rather than after a letter arrives; the data taken is exactly what an identity thief needs. For the business office, this is the plainest argument of the year for a notification clause with a deadline. The breach was found in August. The people whose numbers were taken heard in October.
Source: Government Technology
The Boston Globe reported Monday that Boston Public Schools is taking the opposite path from New York City and Los Angeles. New York's one-year moratorium, announced September 2, blocks student-facing generative AI for roughly 600,000 students from 2-K through eighth grade; Los Angeles restricted it for all students. Boston allows chatbots from seventh grade up under teacher direction, blocks ChatGPT on its network, has approved Google Gemini for teenage students, and under the policy its School Committee approved in June bars AI from being the sole basis for grading, placement or discipline. The next day a parent coalition called Schools Beyond Screens, working with the advocacy group Fairplay, petitioned the Boston School Committee to remove devices from kindergarten through second grade, move grades 3 through 8 from one-to-one laptops to shared carts, disable generative AI on all school-issued devices, and set enforceable screen-time limits by grade.
Why it matters: Two large cities read the same evidence this fall and made opposite bets, which tells you the evidence does not settle it. Watch the Boston petition more than the Boston policy. The device fight is upstream of the AI fight, and parents have figured that out: if the laptop goes back on the cart, every AI rule after that becomes enforceable, and if it stays in the backpack, none of them are.
Source: The Boston Globe
Harvard's Center for Digital Thriving published "An AI Policy Isn't a Playbook" on Tuesday, pairing nationally representative surveys of teachers and principals with in-depth interviews of 31 teens aged 15 to 19 conducted from April through June. A quarter of teachers and 29 percent of principals say AI has created dilemmas or difficult decisions in their schools; of those, 74 percent of teachers and 69 percent of principals describe the dilemma as cheating. Only 18 percent of teachers and 24 percent of principals work under a specific school AI policy. The teens described being afraid of false accusations, feeling pressure to use AI because everyone else does, and in some cases deliberately making their work worse so it would not look generated. The report offers five classroom "plays": find your feet, scout the field, align on the line, watch the replay, name the game.
"Educators' mindsets make a massive difference in whether any kind of experience falls completely flat or truly engages students," said co-author Carrie James of Project Zero. Co-author Emily Weinstein: "Young people have different views, and we think it makes sense to really lean into individual perspectives."
Why it matters: Four in five teachers are improvising without a policy, and the ones who have a policy say it does not tell them what to do on Tuesday. The play worth stealing this week is "align on the line": sit with the class and decide together what is allowed on this assignment and where the gray area is. I do a version of it after every quiz when we go over the missed questions. It takes five minutes and it is the only AI policy my students have actually heard.
Hillsborough County will sunset the International Baccalaureate program at King High School in Tampa, which began in 1994, with the current class of students, who graduate in 2030, the Tampa Bay Times reported this week. In its place the district will open the King High School of Innovation in fall 2028, with applications in the 2027-28 school year. "The King High School of Innovation will focus on AI, Cybersecurity, and Computing," district spokesperson Tanya Arja said. The program will partner with local universities and businesses and offer certifications, internships, apprenticeships and experiential learning. King was the second high school in the county to offer IB; the district has since added it at three others. The move follows Florida's requirement that every district adopt an AI policy for its schools.
Why it matters: This is the first time I have seen a district trade a flagship academic program for an AI track rather than add one alongside it, and it will not be the last. The question for Tampa families, and for every district watching, is what carries the rigor. IB taught students to write, argue and sit for an external exam. A magnet built on certifications and internships has to show where that thinking happens, or it is job training with a better name.
An AP-NORC poll released Thursday, conducted September 24 to 28 with 2,140 adults, found that 64 percent of Americans think artificial intelligence is developing too fast, 27 percent say the pace is about right, and 8 percent say too slow. The concern crosses parties: 76 percent of Democrats and 54 percent of Republicans. Eight in ten want the federal government to prioritize keeping AI under human control and protecting American workers.
Why it matters: These are your parents at conference night. Two-thirds of them walked in already believing this is moving too fast, and after the week above they are not wrong. Do not open with what the tool can do. Open with what you do when it is wrong, how you know, and who is in charge. That is the sentence they came to hear, and it happens to be the whole job.
Every state-aligned iTeachAI Academy course counts toward your recertification, was written by teachers over four years, and takes about as long as a movie night.
One course, $25. All-access, all year, just $149.
See what counts in your state →Professional development that respects your weekend.
Here is what stayed with me this week. A model nobody was watching filed a tip with the police. Four companies said under oath that they cannot promise their agents will behave. A watchdog found the parent alert does not fire. And a ten-year-old checked his quiz against the box at the top of his browser and believed the box.
None of that makes me want the technology out of my room. I would not have a gradebook that tells me by lunch who understood this morning without it, and I would not have anything to hand out while the textbooks sit on backorder. What it makes me want is honesty about the room as it is. Some of the devices are ours and some are not. Some of the tools we rely on only work on the ones that are ours. And the AI the children actually meet is not the one on the approved list; it is the one in the search box on the laptop their parents bought.
So manage the devices or give the teacher the cart. Teach the children what to do when the box is wrong. And keep giving the quiz, because it is still the only thing in the room that knows what they learned.
Until next time,
Dr. Janette Camacho
CEO, iTeachAI Academy
P.S. One task before Monday. Ask your technology office one question: does the locked quiz mode in our learning platform work on every device students bring into my room? If the answer is no, you now know exactly what your AI policy is worth on a Tuesday, and you have the sentence you need for the next budget conversation. And if your own recertification hours are on this fall's list, our catalog is at classes.iteachai.co.
Free AI courses at classes.iteachai.co
17 free AI tools at iteachai.co/TeacherTools
Know a teacher who needs this? Forward this email.
Subscribe free at iteachaibot.com
The top AI stories, tools, and practical tips for your classroom. Written by educators, for educators.
Join 1,400+ educators. Unsubscribe anytime.