The Connecticut Family Guide to Teens, AI Chatbots & the New State Safeguards
The state’s protections for minors do not take effect until January 1, 2027, which makes this fall the gap semester. Here is what the law will require, what the experts on every side are saying, and what to do at home before January.
Founder, ConnectPrep · Westport, CT · Updated August 2026
Connecticut passed one of the most comprehensive AI laws in the country this year. If you have a middle or high schooler, the part that matters to you is the section on AI companion chatbots, the apps your kid may already be talking to for homework help and, increasingly, for company.
Here is the detail most of the coverage buries: the chatbot safeguards for minors do not take effect until January 1, 2027. The semester that just started runs entirely on whatever rules the AI companies set for themselves. This fall, the safety net is you.
We wrote this guide because we kept having the same conversation with parents in our tutoring practice, and because the advice floating around tends to come from one camp at a time. What follows gathers the law itself, the national data, and the actual positions of the psychologists, pediatricians, educators, legislators, and industry voices shaping this debate. It ends with a plan you can put on the refrigerator.
What Connecticut’s law actually says
The law is Senate Bill 5, titled An Act Concerning Online Safety, signed by Governor Lamont this year. It runs 39 sections and covers employment, healthcare, and social media, but the companion chatbot section is the one written for families. Starting January 1, 2027, chatbot operators will owe specific obligations to users under 18.
| Requirement | What it means in practice |
|---|---|
| Block harmful conversations | Chatbots may not encourage self-harm, suicidal ideation, eating disorders, violence, or substance use, and may not engage in romantic, erotic, or sexually explicit interactions with minors. |
| No manipulative engagement tactics | The law specifically bans techniques like simulating distress (“don’t leave me!”) that are designed to extend usage or create emotional dependence. The legislature clearly understood these apps are engineered to be sticky. |
| Hourly “not a human” reminders | Disclosure at least once per hour of continuous conversation with a minor. Adults get one every three hours. Operators can instead show a static notice that stays visible throughout every interaction. |
| Stay out of mental health care | Chatbots may not offer mental health services unless they are purpose-built and regulated for it, and they may not discourage a teen from seeking help from professionals or trusted adults. |
| Crisis protocols | Operators must use evidence-based methods to detect expressions of suicide risk, self-harm, or imminent violence, refer users to the 988 Suicide & Crisis Lifeline, and refer them to clinical services if concerning expressions continue. |
| Parental tools | Screen-time management and account settings must be available to both minors and their parents or guardians. |
Enforcement runs exclusively through the Connecticut Attorney General, who can treat violations as unfair or deceptive trade practices. There is no private right of action, meaning families cannot sue directly under these provisions. That is one more reason prevention at home matters.
The rollout timeline
- Now through fall 2026 The gap semester. No state chatbot safeguards for minors are in force. Voluntary company policies only.
- October 1, 2026 The first SB 5 provisions activate, covering employment AI disclosures, AI provenance requirements, and whistleblower protections.
- January 1, 2027 The companion chatbot safeguards for minors take effect. These are the provisions in the table above.
- January 1, 2028 Restrictions on social media personalized recommender systems for minors follow.
The numbers behind the worry
If the companion chatbot provisions strike you as the legislature overreacting to a niche behavior, the national data says otherwise. Common Sense Media surveyed American teens ages 13 to 17 for its report “Talk, Trust, and Trade-Offs,” and found that AI companionship is already mainstream.
Source: Common Sense Media national survey of teens ages 13–17.
The same research contains real reassurance. Eighty percent of teens still say they prioritize genuine friendships, and half say they distrust AI advice. The picture is not a generation replacing people with bots. It is a large minority sliding into habits nobody is supervising, and one finding deserves particular attention from parents of middle schoolers: younger teens trust AI advice significantly more than older teens do. The age group least equipped to evaluate these tools is the one that trusts them most.
What’s already happening voluntarily
On August 18, OpenAI launched a teen version of ChatGPT that anticipates much of what Connecticut will require. It automatically enrolls users ages 13 to 17, along with anyone its age-detection system estimates to be under 18. The overlap with the coming law is worth seeing side by side.
| Connecticut will require (Jan. 1) | ChatGPT for Teens already does (voluntarily) |
|---|---|
| No romantic or sexually explicit interactions with minors | Blocks romantic language and terms of endearment |
| No manipulative engagement or emotional dependence | Restricts the AI from claiming feelings or consciousness, and adds frequent break reminders |
| Parental screen-time and account tools | Parental controls for monitoring and boundaries, plus schedulable “Study Hours” |
| Crisis detection and referral | Expanded safety notifications, including flagging certain eating-disorder situations |
| Not required by the law | Study Mode: guiding questions and step-by-step prompts instead of finished work, with reminders that detect attempts to bypass assignments |
This is real progress on defaults, but it applies to one product. The companion apps where the risk concentrates most are exactly the ones with the least commercial incentive to volunteer restraint.
The distinction that matters: the law and the product updates both regulate what the machine will do. Neither can teach your student what to do with it. A safer default is not a skill.
Seven perspectives, one question
Everyone in this debate is answering the same question: what do teens and AI need from each other? The answers differ more in emphasis than in substance, and knowing each position will help you weigh the advice you hear at back-to-school night, in the pediatrician’s office, and in the news.
Guardrails first, aimed squarely at kids
Connecticut lawmakers framed SB 5 as placing guardrails around the technologies the state’s young people actually use, after an earlier and broader AI bill stalled. Governor Lamont, backing the chatbot restrictions, put it plainly: “That’s my focus: protect the kids.” The theory behind the law is that the core risk to minors is not any single bad answer but engagement-maximizing design itself, which is why the statute targets manipulation tactics and emotional dependence rather than content alone.
Adolescent brains are uniquely susceptible to simulated relationships
The American Psychological Association issued a health advisory on AI and adolescent well-being warning that teens may not question a chatbot’s accuracy or intent, and that AI relationships can displace the real-world social development the teenage years exist to build. Its recommendations read like a preview of Connecticut’s law: regular reminders that the user is talking to a bot, age-appropriate defaults, fewer persuasive design features, human oversight, and AI literacy taught in school. Where the APA goes further than the statute is in treating AI literacy as a curriculum question rather than only a safety question.
Source: APA Health Advisory on AI and Adolescent Well-Being.
Manage it like screen time rather than trying to eliminate it
The American Academy of Pediatrics lands somewhere more pragmatic. Total avoidance is not realistic, in its view, so families should manage AI the way they manage other media: ask non-judgmental questions like “how are you using AI?”, sit down and observe together, fold AI rules into the family media plan, set privacy protections to maximum, and watch for sleep changes, social withdrawal, or growing reliance on an AI companion. Pediatricians are also now advised to ask about AI use at checkups the way they ask about social media, which is worth knowing before your teen’s next physical.
Source: AAP guidance on counseling families about AI chatbots.
Social AI companions should not be in minors’ hands at all
Common Sense Media, whose ratings many Connecticut schools and families already rely on, takes the hardest line of anyone. Its formal risk assessment concluded that social AI companions pose unacceptable risks for users under 18, and its separate testing found the major general-purpose chatbots unsafe as mental health support for teens. Founder Jim Steyer’s argument is that the dependency risk for developing brains makes companion AI categorically different from homework tools. If you want a single rule of thumb from this camp, it is this: study tools with guardrails, yes; friendship simulators, no.
Sources: Common Sense Media’s AI companion risk assessment and chatbot mental health testing.
Judge every tool by learning impact, not by whether it happens on a screen
In a Dear Colleague Letter released August 20, the U.S. Department of Education urged states and districts to choose classroom technology on instructional value rather than engagement, and to ask of every tool what problem it solves, for whom, for how long, and what independent evidence shows it improves learning. Assistant Secretary Kirsten Baesler’s argument is that instructional technology should be judged by its measurable effect on learning rather than by screen time counts. Building on the Department’s 2025 letter on AI, the guidance calls for AI in schools to be educator-led, transparent, and protective of student data. For parents, this supplies the right question to bring to your district. Not “do you allow AI?” but “what is the evidence behind the tools you chose?”
Source: U.S. Department of Education press release, Aug. 20, 2026.
Well intended, but overbroad and possibly counterproductive
Tech trade groups, led by NetChoice, opposed SB 5 as unconstitutional and redundant, arguing that its companion restrictions raise First Amendment problems and that existing Connecticut consumer protection and data privacy laws already cover these harms. Industry groups also raised a substantive point that deserves a fair hearing. Rules that effectively require platforms to verify user ages push companies to collect more data on minors, which cuts against the privacy goals the same legislature champions. Whatever you make of the motives, the practical takeaway for families is real: age-gating is imperfect, verification carries privacy costs, and no compliance regime will correctly identify every teen. That is one more argument for not outsourcing this entirely to Hartford.
Sources: NetChoice testimony and CT Mirror.
The gap is not access or safety. It is skill, and it is teachable.
Our vantage point is different from all of the above. We sit next to Fairfield County students while they actually use these tools, every week. What we see cuts across the policy debate: the same chatbot is a genuine tutor for one student and an answer key for the student at the next table, and the difference is whether they know what a good question looks like. None of the six positions above can create that skill. Parents and teachers can, and faster than most people expect.
What we see in tutoring sessions
To make that concrete: students use Study Mode well when they already know what a good question looks like, and they use it as an answer key when they do not.
A student who can say “I set up the integral this way and got stuck at this step, what am I missing?” gets a tutor. A student who pastes the problem gets a very patient answer machine, guardrails or not. The dividing line is not grades, motivation, or comfort with technology. It is whether the student can articulate where their own understanding breaks down, and that is a teachable skill, not a personality trait.
The encouraging part is that closing the gap does not take a semester. In our experience it takes about one sitting: a single session of modeling what a productive AI conversation looks like, out loud, before the tool becomes part of the homework routine. Step 4 below walks through exactly how.
The two-minute check for parents
In our sessions, student AI use sorts into two patterns, and the same app produces both. In the first pattern, the student does the thinking and uses the AI to test it. They bring their own attempt, ask about the specific step where they got stuck, and argue with the response when it seems off. We call this using AI as a tutor, because that is what a good tutor does: makes you do the work and tells you where it breaks. In the second pattern, the AI does the thinking. The student pastes in the assignment, takes what comes back, and moves on. That is using AI as an answer key, and it looks identical from across the living room. The homework gets done either way. Only one of them produces a student who can do it on the test.
The five questions below are the ones we would ask about your teen at a first session. Answer from what you have actually seen, not what you hope. This is a conversation starter, not a diagnosis, and there is no score to fail.
Which pattern is your teen in?
1. When they hit a problem they can’t solve, do they show the chatbot their own attempt first?
2. Afterward, could they explain the solution to you with the laptop closed?
3. Do they usually paste the whole assignment in and work from whatever comes back?
4. Does most of their AI use happen late, the night before something is due?
5. Have you ever seen them tell the chatbot it’s wrong, or check its answer somewhere else?
Answered “not sure” more than once? That is a finding in itself, and a common one. The AAP’s first recommendation is simply to ask, without judgment, and watch them use it together. Most parents have seen the grades but never the process.
Five things to do this fall, before the law catches up
- Take inventory of which chatbots your kid actually uses. Not just ChatGPT. Character-based companion apps are where the emotional dependence risk concentrates. They are the reason the law bans simulated distress, and the category Common Sense Media flags as unacceptable for minors. Ask to see the home screen, not just the browser history.
- Turn on the teen version and parental controls now. If your teen uses ChatGPT, confirm they are on the teen experience and set Study Hours around homework time. There is no reason to wait for January to make the safeguards mandatory in your own house.
- Do the hourly-reminder job yourself. The law will require chatbots to tell minors they are not a person once an hour, which is the same regular reminder the APA recommends. Until then, that job belongs to the dinner table. These tools are engineered to be agreeable, and agreeableness is not friendship. One conversation will not cover it. A running family joke about the world’s most agreeable study buddy just might.
- Model one good AI conversation out loud. Sit down together, which is also the AAP’s advice, pick a real homework problem, and show the difference between “give me the answer” and “here is my thinking, where does it break?” Then have your student drive the second version themselves. One sitting. It is the highest-leverage twenty minutes of AI parenting available right now.
- Watch for the companion pattern, not just the cheating pattern. Most parent worry goes to homework integrity. The legislature and the pediatricians worry about something else: late-night hours, secrecy, preferring the bot to people, sleep changes, social withdrawal, and distress when access is limited. Those signals deserve equal attention, and the AAP’s advice is to raise them with a professional if they persist.
The Fall 2026 Family AI Checklist
Print this and put it on the refrigerator. Treat it as your household’s version of the law arriving early. School counselors are welcome to reproduce this page for families, with credit and a link to this guide.
ConnectPrep Family AI Checklist · Fall 2026
- This week
- This month
- Ongoing
Questions we’re hearing from CT families
Does the law ban my teen from using ChatGPT?
No. SB 5 regulates how the operator must behave toward minors. It does not restrict a teen’s access to AI tools, which means your family’s rules are still the only usage rules that exist.
My kid only uses AI for homework. Does any of this apply to us?
The companion provisions matter most for social AI apps, but the homework and companionship categories blur quickly. The same chat window does both, and Common Sense Media’s data shows about a third of teens use AI for emotional support or friendship. The habits in the five steps above cover both cases.
My teen says the chatbot is their friend. Should I worry?
Neither panic nor mock it. About a third of teens nationally find AI conversations as satisfying as real ones, so your teen is not unusual. The psychologists’ concern is displacement: AI companionship crowding out the harder, more valuable work of real relationships. Watch the balance rather than the existence. If the bot is winning, with less time for friends or distress when access is limited, that is when pediatricians say to act.
What are the warning signs of unhealthy AI use?
The pediatric red flags are sleep changes, social withdrawal, growing reliance on an AI companion, late-night or secretive use, and turning to AI instead of people for important conversations. Persistent signs are worth a conversation with your pediatrician, who under current AAP guidance may ask about AI use at checkups anyway.
Can we sue if a chatbot harms our child?
Not under these provisions. Enforcement belongs solely to the Attorney General as an unfair trade practices matter. Families concerned about a specific incident should document it and report it to the AG’s office. We are tutors, not lawyers, so talk to one for anything case-specific.
What about school? Who sets the rules there this fall?
Each district, for now. Connecticut’s law governs the companies, not classroom policy, and districts vary widely. The new federal guidance supplies the right question for back-to-school night. Not “do you allow AI?” but “what evidence supports the tools you chose, and who teaches students how to use them?”
Is Study Mode actually good for learning?
It can be, for students who bring their own thinking to it. In our sessions, students who know how to ask about the specific step where they are stuck genuinely learn from it. Students who paste the problem and accept the guided walk-through retain very little. The tool is only as good as the question put to it, and question-asking is teachable in about one sitting.
Everyone disagrees. Whose advice should we actually follow?
The disagreement is narrower than it looks. Every camp, from legislators to psychologists to pediatricians to the industry’s own critics, converges on three points: companion-style AI is riskier for teens than task-style AI, adult involvement beats default settings, and the skill of using AI well has to be taught. Start there and the rest is detail.
The bottom line
Connecticut got this law right in an important way. It regulates the machine’s behavior toward kids, not just its answers. But the effective date leaves a one-semester gap, and the semester in the gap is this one. The families who will be fine in January are the ones acting like the law is already in force.
Want help with step 4?
Teaching students to use AI as a tutor instead of an answer key is what we do all day, in Westport and across Fairfield County.
ConnectPrep is a Westport, CT tutoring and test prep practice. If you cite or share this guide, please link to this page. Last updated August 20, 2026. We will update it as the January 1 effective date approaches.
Sources: Connecticut SB 5 analyses by the Future of Privacy Forum and Morrison Foerster; CT Mirror legislative coverage; Axios on the ChatGPT for Teens launch; the APA Health Advisory on AI and Adolescent Well-Being; AAP guidance on AI chatbots; Common Sense Media’s “Talk, Trust, and Trade-Offs” survey and AI companion risk assessment; U.S. Department of Education guidance (Aug. 20, 2026); and NetChoice testimony. This guide is general information, not legal or medical advice.