When 44% of Teens Get Blocked: Rules That Actually Survive

Common Sense Media’s August 2026 survey produced the single most useful number for any parent writing household technology rules: among teens who use AI for schoolwork, “44% have had an AI tool blocked on a school network or device, and 59% of them just switched to a personal device.” The block worked technically. It failed practically. Which means the question for AI rules for teens with a personal device isn’t how to build a better wall. It’s what kind of rule keeps working when the wall is beside the point.

Key Takeaways

• Common Sense Media (Aug 18, 2026): 44% of teen AI users have had a tool blocked at school; 59% of those switched to a personal device. Only 27% say a teacher ever discussed safe AI use.
• Control-based rules fail predictably when a second device exists. Trust-based rules fail differently: they degrade slowly and recover after a violation.
• The evidence supports moderation over prohibition. PISA 2025 found moderate AI users outscored both non-users and near-daily users, and students taught to assess AI output scored slightly higher.
• OpenAI’s design encodes this: parents get settings and Quiet Hours but explicitly cannot read their teen’s chats.
• The durable rules are the ones with a visible outcome check, not a blocked input.

Why control-based rules break at the device boundary

A control-based rule works by preventing access: a network block, an app restriction, a device lock. Its failure mode is simple and it has now been measured. When a school blocked an AI tool, 44% of teen AI users hit that block, and 59% of them switched to a personal device. Four out of seven.

The relocated use is worse than the original use in every respect. It’s unsupervised. It happens on a consumer account instead of a school account with FERPA-aligned terms. Nobody is teaching evaluation, and Common Sense found only 27% of teens report a teacher ever discussing safe AI use, so the 73% who weren’t taught are now using it entirely alone.

Compare that to what the outcome data actually rewards. PISA 2025, released September 8, 2026, found that moderate AI users, roughly once a month to once or twice a week, scored higher in science than both rare users and near-daily users. And students who reported learning at school how to assess AI-generated information tended to score slightly higher, an effect strongest among frequent users. The OECD is explicit that these are associations.

So the thing that correlates with better outcomes is supervised, moderate, evaluated use. Control-based rules produce unsupervised, unevaluated use as their failure mode. That’s not a small mismatch.

None of which means controls are worthless. For a nine-year-old with no second device, a control-based rule is the whole plan and it works. The problem is specific to teenagers with their own phone, which is to say, most of them.

The durability table

Each row is a real approach. “Durability” is my assessment of how long it survives contact with a motivated fifteen-year-old who owns a phone.

Approach	How it works	Durability	Fails when	What it teaches
Network block	Router or DNS blocks AI domains	Days	They use cellular data	That your rules are a technical obstacle
App restriction on the family device	Parental controls block the app	Weeks	A web version, or a friend’s phone	That the specific app is forbidden, not the behavior
Full ban, all devices	Prohibition with consequences	Weeks to months	First unsupervised evening	That AI use is a secret to keep
Time cap	X minutes of AI per day	Months	The cap doesn’t distinguish study mode from answer mode	That quantity is what matters
Quiet Hours plus Study Hours	Platform-level schedule (OpenAI offers both)	Months	They use a different tool	That there’s a time and a mode for this
Task-based rule	Named allowed and disallowed tasks	Long	Never fully; it degrades at the edges	That the task determines what’s appropriate
Outcome check	Two-minute oral explanation after work	Longest	Only if you stop doing it	That understanding is what ends homework
Written joint policy	Teen drafts, parent edits, both sign	Long	If it was never really joint	That they have standing in the rule

The two bottom-half approaches share a structural feature: they don’t depend on blocking anything. A task-based rule and an outcome check both work identically whether the kid is on your wifi, their cellular data, or a friend’s laptop. That’s what makes them durable. They’re not harder to evade; they’re not evadable in the same way, because the thing being checked is on the output side.

What a trust-based rule actually looks like

“Trust-based” is a term that invites eye-rolling, so let me be concrete. It does not mean no rules. It means rules whose enforcement mechanism is a check you perform on the work rather than a lock you place on a tool.

The task list. Written, specific, with a generous yes-column. Allowed: explaining a concept you’re stuck on, quizzing you from your own notes, checking a computation after you’ve done it, summarizing a reading you’ve already read. Not allowed: drafting graded writing, doing a problem set, generating a summary of something you haven’t read. The PISA data supports this shape directly: drafting showed the sharpest negative gradient, 509 in science for students who rarely used AI to draft versus 481 for near-daily use.

The outcome check. Two minutes, every time, on the work itself. “Walk me through this part.” This is the University of Chicago Law School’s mechanism, scaled down: its July 9, 2026 AI strategy replaced detection software with mandatory oral defenses where students answer questions probing “the paper’s reasoning and argument implications.” It cannot be evaded by switching devices.

The disclosure norm. One line at the end: what you used, for which part. Katy ISD requires students to “verify AI-generated information and disclose or cite AI use when appropriate.” Adopting it at home makes disclosure ordinary rather than an admission.

The platform settings, as support not scaffolding. Turn on OpenAI’s Study Mode and Study Hours, set Quiet Hours, turn off “improve the model for everyone.” Note that OpenAI states parental controls “do not let a parent or guardian read or monitor the teen’s conversations,” which tells you the platform itself is designed around visibility rather than surveillance.

A stated recovery path. What happens after a violation, agreed in advance. This is the piece most households skip, and it’s why control-based rules collapse permanently after the first breach: there’s no defined way back.

Running it without becoming the enforcement arm

Write the policy together, and let them win two points

A rule a teen helped draft gets followed at a rate an imposed rule doesn’t. Give them real input on at least two items and accept an answer you don’t love. You’re buying compliance on the other eight, and you’re also finding out what they actually believe about the tool.

Make the outcome check unconditional

Every assignment, not just suspicious ones. If the check only appears when you doubt them, it’s an accusation. If it happens always, it’s just how homework ends. This is also the single most evidence-backed study activity in the whole system: Roediger and Karpicke (2006, Psychological Science) found repeated testing produced “substantially greater retention than studying.”

Say out loud that you can’t monitor the chats

Counterintuitive, but it works. OpenAI’s own documentation says linking “does not give a parent or guardian access to a teen’s conversations, chat history, or real-time activity.” Telling your teen you know this, and that the system therefore rests on their account of their own work, reframes the whole arrangement. Pretending you can see more than you can is how you lose credibility in one round.

Define the recovery path before you need it

One sentence: if the rule gets broken, here’s the conversation, here’s the temporary change, here’s how it goes back to normal. Without this, a violation becomes a cliff, and a kid facing a cliff hides everything. With it, you get told about problems.

Keep the yes-list longer than the no-list

Three yeses for every no. This isn’t softness; it’s about making the rule usable. A policy that reads as a list of prohibitions gets treated as an obstacle course. A policy where most legitimate uses are explicitly permitted gets treated as a guide.

What not to do

Don’t install monitoring software and then also claim to be running a trust-based system. Pick one. The hybrid teaches your teen that your stated principles and your actual behavior differ, which is a much more expensive lesson than anything they’d learn from an AI chatbot.

What to Watch For Over the Next 3 Months

• Week 4: Is the outcome check still happening every time, or only when you’re suspicious? The drift toward selective checking is what turns it into an accusation.
• Month 2 red flags: A second account or tool you didn’t know about. Treat it as information about the rule, not just about the kid: something in the task list was too restrictive to be usable.
• Month 3 self-check: Ask your teen what the household AI rule is. If they can state it from memory, it’s a real rule. If they can’t, it’s a document.

Frequently Asked Questions

What percentage of teens get around school AI blocks?

Common Sense Media’s August 2026 survey found that among teen AI users, “44% have had an AI tool blocked on a school network or device, and 59% of them just switched to a personal device.” So most blocked teens continued using AI, just unsupervised.

Should I block AI apps on my teen’s phone?

For a younger teen without alternatives, it can work. For an older teen with their own phone and data, blocking mostly relocates the behavior. The evidence favors named-task rules plus an outcome check, which work regardless of which device they’re on.

Isn’t a trust-based approach just giving up?

No. It replaces one enforcement mechanism with another. Instead of blocking the input, you check the output: a two-minute oral explanation of the work. The University of Chicago Law School made exactly this switch in July 2026, dropping detection software for mandatory oral defenses.

Can I read my teen’s AI conversations?

Not on ChatGPT. OpenAI states that linking accounts “does not give a parent or guardian access to a teen’s conversations, chat history, or real-time activity.” You get settings control, Quiet Hours, and limited safety notifications.

What’s the most important rule to write down?

The task list, with specific allowed and disallowed uses. PISA 2025 data shows the tasks differ sharply in impact: drafting writing assignments showed a 509-versus-481 science score gap between rare and near-daily users, while summarizing assigned reading was safer at weekly frequency.

What do I do the first time a rule gets broken?

Use the recovery path you defined in advance: a conversation about what happened, a temporary adjustment, and a stated route back to normal. Households without a defined recovery path tend to escalate to a total ban, which is the approach the data says relocates rather than reduces use.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Common Sense Media. (Aug 18, 2026). “Teens in the AI Era: Schoolwork and Skills That Matter.” https://www.commonsensemedia.org/research/teens-in-the-ai-era-schoolwork-and-skills-that-matter
2. OECD. (Sept 8, 2026). “PISA 2025 Results (Volume I).” https://www.oecd.org/en/publications/pisa-2025-results-volume-i_73451bc5-en.html
3. Fortune. (Sept 11, 2026). “Students are using AI in school and test scores are down.” https://fortune.com/2026/09/11/students-ai-school-test-scores-moderate-use-vs-daily-users/
4. OpenAI Help Center. (2026). “Parental controls in ChatGPT: FAQ.” https://help.openai.com/en/articles/12315553
5. University of Chicago Law School. (July 9, 2026). “Rethinking Legal Education in the AI Era.” https://www.law.uchicago.edu/news/ai-strategy-statement
6. Roediger, H. L., III, & Karpicke, J. D. (2006). “Test-enhanced learning: Taking memory tests improves long-term retention.” Psychological Science, 17(3). https://doi.org/10.1111/j.1467-9280.2006.01693.x
7. Covering Katy News. (Aug 11, 2026). “Katy ISD’s new AI rules: What parents and students need to know.” https://coveringkaty.com/education/katy-isd-s-new-ai-rules-what-parents-and-students-need-to-kn/
8. NPR/Ipsos. (June 5, 2026). “Poll: Teachers worry AI is impacting students’ critical thinking.” https://www.npr.org/2026/06/05/nx-s1-5779757/school-ai-education-students-teachers-poll-critical-thinking
9. Apple Newsroom. (June 8, 2026). “Apple unveils next generation of Apple Intelligence, Siri AI, and more.” https://www.apple.com/newsroom/2026/06/apple-unveils-next-generation-of-apple-intelligence-siri-ai-and-more/

Related reading: when school AI blocks meet personal devices, family screen time agreements and what research says works, and the home AI policy one-page template.