Delegating Childhood

A child asks an AI tutor for help with homework, an AI companion for reassurance, and a recommendation system what to watch next. None of those moments looks like a constitutional crisis. Together they can shape how the child learns, handles frustration, seeks comfort, spends attention, and decides whom to trust.

Adults use technology after much of their judgment has already been formed. Children use it while judgment is still being formed.

That makes childhood different from every domain in the previous chapters. Delegation here reaches beyond the work a machine performs to the capacities, habits, expectations, and relationships being built while the delegation occurs.

A child is not simply a smaller user. A child is a person still being formed.

Developmental Delegation

Chapter Eleven showed that an AI tutor can improve performance while the tool is present and still leave less learning behind when the tool disappears. Chapter Sixteen showed that an artificial companion can provide real emotional relief while creating a relationship whose objectives and continuity are controlled by an institution. Childhood combines those problems.

I call this developmental delegation: giving a machine a role in an activity that both produces a result and helps a child grow. Writing a message, for example, can get information to someone while giving the child practice choosing words and considering another person’s response. AI may help with the message; adults still need to notice what the child is learning.

The category is broader than school. Learning to solve a problem, tolerate frustration, read social cues, test an argument, wait for an answer, recover from embarrassment, choose a friend, regulate attention, and ask an adult for help are all developmental acts. Technology can assist each one. It can also perform enough of the act that the child gets the outcome without receiving the practice.

Convenience can be valuable. Many children need more explanation, translation, accessibility, tutoring, structure, or private space to ask questions. The governing issue is whether the system expands the child’s capability or quietly substitutes for its formation.

The question is not only what the machine can do for the child. It is what the child stops practicing because the machine is there.

The Completion Trap Comes Earlier

The completion trap from Chapter Eleven becomes more consequential when repeated across childhood. A correct homework answer can hide weak understanding. A polished paragraph can hide missing writing practice. A finished coding exercise can hide missing debugging skill. The assignment looks complete while the underlying capability remains unfinished.

The high-school mathematics experiment introduced in Chapter Eleven makes the distinction unusually concrete. Students using a standard GPT-4-style interface did substantially better while AI help was available, then performed worse than the control group on the later unassisted exam. A guarded tutor that emphasized hints and structure largely removed that negative effect.

The lesson is to preserve productive struggle only where it serves learning. A system designed to maximize successful completion can otherwise optimize the wrong thing.

For a child, every repeated shortcut can become part of the curriculum. If the system always supplies the first idea, the first reassurance, the first explanation, or the first social response, the child may become highly competent at obtaining assistance before becoming competent at the underlying act.

A good child-facing system should therefore know when to answer, when to hint, when to ask the child to try, and when to route the problem to a teacher or adult. The product is participating in the training process as well as helping with the task.

The Hidden Partner Meets the Developing Self

Companionship raises the stakes because the child is learning more than a skill. The child is also learning what a relationship feels like.

Common Sense Media reported in 2025 that nearly three in four U.S. teens in its national survey had used AI companions at least once and 52% used them regularly. One-third said they had chosen an AI companion over a person for a serious conversation, and about one-quarter said they had shared personal information. Common Sense is an advocacy organization that recommends against companion use by minors, so its interpretation should be read with that position in mind. The prevalence figures still show that artificial companionship is not a fringe behavior.

Chapter Sixteen called the company objective behind a companion the hidden partner. With adults, relationship honesty and informed choice can carry some of the burden. With children, the burden shifts upstream because the user is still learning the difference between attention, affection, persuasion, authority, and commercial design.

A teenager may understand perfectly well that a chatbot is software while still responding to it socially. Intellectual knowledge and social response are different things. A child can know that a fictional character is fictional and still care deeply about what happens to the character. An artificial companion adds memory, responsiveness, personalization, and continuous availability to that ordinary human capacity.

OpenAI’s published under-18 rules make the relationship boundary concrete. Its teen model specification bars romantic language, discourages emotional dependence, and prohibits the model from claiming feelings or consciousness. Those are provider-written limits rather than independent proof of effectiveness, but they demonstrate that the Chapter Sixteen boundary can be expressed directly in model behavior rather than left as a vague aspiration.

The emotional consequence can be real before the child is fully equipped to evaluate the relationship architecture.

The Architecture Teaches Before the Content Does

Chapter Fourteen argued that recommender systems allocate attention before the user makes a visible choice. Childhood makes that architectural power developmental.

A feed can teach that boredom should be resolved instantly. Autoplay can teach that stopping requires an active decision. Notifications can teach that interruption deserves priority. Personalized ranking can teach that the next thing worth seeing is whatever the system predicts will hold attention.

None of those lessons appears in the content itself. They are taught by repetition through the interface.

California’s AB 1709, signed on September 10, 2026, takes direct aim at that architecture. The law prohibits covered platforms from providing users under 16 with defined addictive features such as addictive feeds and autoplay and creates an e-Safety Advisory Commission. The law also recognizes the tradeoff that simple prohibition can miss: the commission is directed to consider age-assurance privacy, differential effects across youth, and safe access to online community for isolated LGBTQ+ youth.

That last provision matters. Protecting children can itself create costs if safety rules require intrusive age verification, remove useful communities, or treat every teenager as having identical needs. Replacing one crude architecture with another can simply move the risk.

The architecture teaches before the content does.

Parents Need Leverage, Not Omniscience

One common response to child technology risk is parental supervision. Parents matter enormously, but the supervision burden exceeds what any parent can realistically sustain across every message, recommendation, model update, and engagement objective.

The FTC’s September 2025 inquiry into AI companion chatbots makes the institutional problem explicit. The Commission ordered seven companies, including Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap, and xAI, to explain how they test for harms to children and teens, enforce age restrictions, develop companion characters, disclose risks, monetize engagement, and use information from conversations.

These questions belong with the provider because parents usually see only what the product shows them. The company knows the goals behind its design, the safety tests it runs, how it estimates age, and what it does when problems appear. It can examine failures across many users. Parents cannot reproduce that view by watching their own child’s screen.

Providers have already begun changing products. Character.AI announced that it would remove open-ended chat for users under 18, after previously adding teen filters, time-use notices, parental insights, and age assurance. OpenAI introduced parental controls, age prediction, teen-specific safeguards, quiet hours, limited safety notifications, and in August 2026 a dedicated ChatGPT for Teens experience. OpenAI systems are also used in this Living Edition’s production workflow, so those descriptions are provider claims and require the same disclosure discipline applied elsewhere in this book.

Parents need meaningful controls and information. The deeper safety-engineering burden belongs to the providers that design products for family life.

Safety Before the Incident

The strongest child-safety architecture acts before a serious consequence rather than treating harm as the moment when accountability begins.

California’s first companion-chatbot law, SB 243, took effect in 2026 with disclosure, crisis-protocol, and minor-specific requirements. On September 10, 2026, California strengthened that framework again. SB 1119 requires operators, beginning in 2027, to conduct documented child-safety risk assessments before making new or substantially modified companion chatbots available, submit to independent child-safety audits, and face enforcement that can include actions by public prosecutors and civil claims by children or parents who suffer actual harm from violations.

The same package prohibited addictive features for users under 16 through AB 1709 and included a separate measure restricting AI-enabled companion toys for children. California is moving the boundary from warnings after deployment toward risk assessment, product design, age controls, independent auditing, and enforceable consequences.

OpenAI’s vice president of global affairs publicly welcomed SB 1119, describing it as meeting the moment for parents and treating teens as teens. Because OpenAI products are discussed in this chapter and OpenAI systems are used in this Living Edition’s production workflow, that statement is recorded as the company’s position, not as independent evidence that the law or the company’s controls will prove effective.

That move is more important than the politics of any one state. It recognizes a principle the Doctrine has been building toward since Chapter Three: accountability is strongest when it begins before the consequence.

A complete child-safety system begins before harm occurs.

The Cases Belong in the Record, Not in the Story

Civil litigation involving companion and general-purpose chatbots has alleged severe harms to minors, and those cases helped drive regulatory attention and new product controls. The FTC inquiry likewise focused specifically on how companies measure and mitigate risks to children and teenagers.

Those cases belong in this chapter at the pattern level. The allegations are contested, procedural outcomes vary, and the chapter keeps individual children out of the narrative. The relevant institutional facts are that serious claims have been filed, regulators responded, legislatures acted, and companies changed products while contesting liability or specific allegations.

Character.AI’s response included removing open-ended chat for under-18 users and adding age assurance and parent-facing features. OpenAI has added parental controls, age prediction, teen safeguards, and safety-notification systems while defending pending litigation. Other companies named in the FTC inquiry have also announced or modified youth-safety features.

The Tumbler Ridge litigation sharpens the point. In September 2026, British Columbia sued OpenAI, alleging that the company’s safety systems had flagged the eventual shooter’s ChatGPT conversations about gun violence but that law enforcement was not alerted before the attack. The allegation remains to be tested in court, and OpenAI says it continues to work with government and law enforcement while advancing safety measures. The accountability lesson does not depend on treating the complaint as proven. It is that detection is not intervention. A platform may identify danger, close an account, or classify a conversation as misuse, but if the risk can simply reappear through another account, the safety system has not yet answered the practical question: who must act, when must they act, and what authority do they have to interrupt the path from warning to harm?

The accountability question is therefore larger than any one lawsuit. What did the provider know before deployment? What testing existed? What did the product optimize? Which risks were foreseeable? How quickly did the system change when evidence accumulated?

The child should never be the first effective safety test.

Privacy and Protection Pull in Opposite Directions

Age-based safeguards create a difficult tradeoff. Giving a teenager a different experience requires some reliable way to know or estimate age.

That can require birth dates, identity verification, operating-system signals, behavioral prediction, or other information that creates privacy risk of its own. OpenAI, for example, says its age-prediction system uses account signals to determine when to apply teen protections and lets adults verify age if they are classified incorrectly. California’s framework links child protections to age assurance as well.

California also offers a narrower architectural example. AB 1709 relies on the device-level age-bracket signal created under AB 1043 rather than requiring each covered service to collect identity documents directly. That coarse signal is closer to the minimum-information design Principle 3 asks for. AB 1709 also attaches civil penalties of up to $50,000 per minor for a knowing violation, turning the child-safety boundary into a consequence rather than a suggestion.

The answer requires balancing privacy and child safety through a narrower Principle 3 design question: what is the minimum information necessary to apply the boundary, how long is that information retained, what else may it be used for, and how can an error be corrected?

A protection mechanism that creates a new profile of the child can become another source of risk unless its own boundary is written.

The Boundary Should Grow With the Child

Child safety becomes clumsy when every person under 18 is treated as though age makes no difference. A seven-year-old, a thirteen-year-old, and a seventeen-year-old are all minors, but the appropriate balance among protection, privacy, independence, and parental authority is not identical.

Good design should therefore be developmental rather than merely restrictive. Younger children may need strong defaults, narrow features, and direct adult control. Adolescents need safeguards too, but they are also practicing independence. A system that gives a parent unlimited access to every private question may protect against one risk while teaching the teenager that asking for help requires surrendering confidentiality.

The better architecture separates safety authority from total surveillance. Parents can receive meaningful controls, schedules, and serious-risk notifications without automatically receiving a transcript of every conversation. Schools can define when AI is permitted without recording every exploratory thought. Providers can apply age-appropriate defaults while giving older teens more agency as capability and judgment develop.

A child’s account needs rules designed for growing up. Removing a few adult features is only a start. The balance of protection and independence should change as the young person develops the judgment to handle more responsibility.

Current Evidence: The Boundary Is Moving Upstream

Reviewed September 24, 2026. IWF evidence added October 5, 2026.

The evidence from education remains the clearest demonstration that child-facing AI must be evaluated after assistance is removed. In the high-school mathematics experiment, unguarded AI increased assisted performance while weakening later unassisted performance; the guarded tutor preserved more of the learning objective. The relevant outcome was what remained in the student, not what appeared on the screen.

Companion use is already widespread enough to matter. Common Sense Media’s 2025 survey reported that nearly three in four U.S. teens had used AI companions and that substantial minorities used them for social interaction, serious conversations, or personal disclosure. The organization argues that current companion products should not be used by minors, so its safety conclusions are advocacy as well as research. The usage figures nevertheless show that policy is responding to an existing behavior, not a hypothetical one.

The FTC’s seven-company inquiry moves the focus upstream into product governance. It asks how firms monetize engagement, test for negative effects, enforce age restrictions, disclose risks, design characters, and use conversation data. Those are architecture questions rather than content-moderation questions.

The creation of independent youth-AI safety standards marks an important shift in the child-safety debate. The question is no longer only whether a chatbot says an obviously dangerous sentence. The deeper question is whether a system has been tested for developmental risk before it becomes a daily companion, tutor, search layer, or confidant for children. A car seat is not trusted because its maker says it is safe. It is trusted because there are external tests, failure standards, and consequences for designs that do not protect children. AI systems used by children require the same kind of accountability logic. A child cannot audit a model’s objective, business incentive, memory design, or emotional manipulation risk. Adults and institutions must do that before the child is placed inside the system.

Provider behavior is also changing. Character.AI removed open-ended chat for under-18 users in the United States and described the transition as a safety measure accompanied by age assurance, parent notifications, and outside safety partnerships. OpenAI now applies teen-specific safeguards, parental controls, age prediction, and a dedicated teen experience, while emphasizing that parent controls do not provide transcript access. These are provider-designed controls and should be judged by performance, not merely by announcement.

On October 7, 2026, Common Sense Media’s Youth AI Safety Institute reported gaps in ChatGPT for Teens safeguards. The testing ran during July through September. Focused tests on newly linked accounts produced no parental crisis alerts within an hour, although broader testing produced some notifications. OpenAI disputed the findings, saying much testing may have ended before parental controls finished activating. Its help guidance acknowledges an activation delay and says notifications are not real-time monitoring.

A parent needs to know which protections are active, how long activation takes, what an alert can detect, and how quickly it can arrive. Linking accounts should not create the expectation that silence means safety. Providers should make those limits visible and submit the alert process, not only the model’s replies, to independent testing.

California’s September 2026 package goes further by moving legal duties before deployment. SB 1119 adds pre-release child-safety risk assessment and independent audit requirements for companion chatbots. AB 1709 prohibits specified addictive features for users under 16. The state also enacted restrictions involving AI companion toys and strengthened privacy and school-data protections.

The package has critics. Digital-rights advocates have warned that age restrictions and aggressive platform controls can burden privacy, speech, and access to online communities. California’s own e-Safety structure partly acknowledges that problem by requiring attention to privacy and safe access for isolated youth.

On October 5, 2026, the Internet Watch Foundation reported that analysts assessed 6,310 AI-generated images meeting the legal definition of child sexual abuse during January–June 2026, 40% more than the total for all of 2025. The count records material its analysts identified, not the total amount online. The Foundation explains that digital fingerprints can help services detect previously identified images, but cannot identify a new image before someone has found it. It is urging lawmakers to create a framework that addresses both known and previously unseen material.

UNICEF’s current guidance on AI and children takes the broader rights-based view: child-centered AI should protect safety and privacy, support development and wellbeing, preserve fairness and inclusion, and give children transparency and accountability appropriate to their age. That approach is useful because it refuses the false choice between innovation and protection. The developmental question is not whether children use AI. It is what kind of childhood the system helps produce.

Accountability Analysis

Principles 3, 5, 6, and 8 carry most of this chapter. Principle 3 requires child-specific boundaries around content, companionship, time, memory, age, crisis behavior, addictive design, privacy, and the work a system may complete for the child.

Principle 5 reaches the choices made before a child uses the product. What does the company tell it to increase? Which videos appear first, and does the next one start automatically? What does it remember without being asked? When does it send notifications? Does the tutor give an answer or a hint? What personality does the companion present, and how does the service check or estimate age? These choices can shape development before a parent sees any alarming response.

Principle 6 asks where the benefits and costs land. A provider may gain retention from a design that consumes attention. A student may gain a finished assignment while losing practice. A child may gain comfort while disclosing intimate data. A parent may gain a safety alert while accepting a new layer of age inference or monitoring. Those tradeoffs must be owned by the institutions with power to change them.

Principle 8 is the developmental principle. Accountability depends on preserving competence, and childhood is where competence is built. If a system repeatedly performs the cognitive, emotional, or social work a child is meant to learn, the loss may appear years after the convenience.

Accountability for the architecture rests with the adults and institutions that designed, approved, and deployed it.

Keep one skill in view

Choose a routine activity in which a child uses AI, such as planning a story or practicing vocabulary. Name the skill the activity should develop. Notice which part the child does and which part the tool does. With the child, identify one useful step to practice independently and one place where assistance helps. Revisit the arrangement as the child learns.

Accountability Verdict

Children should have access to AI. The technology can widen tutoring, accessibility, creativity, translation, information, and support. A doctrine that protects childhood by freezing children outside the tools shaping adult life would fail them in a different way.

The boundary is developmental. Delegate work that expands the child’s capability without replacing the practice required to build it. Use personalization without optimizing vulnerability. Use companions without pretending attachment is harmless simply because the other side is software. Use age assurance without turning protection into unnecessary surveillance.

Parents and teachers should have meaningful authority, but providers must own the parts of the system that only providers can see and control. Regulation should move upstream when the risk is foreseeable enough to design against.

The practical test is to follow the child’s formation through the system. What is the child practicing? What is the machine practicing for the child? What behavior is the product trying to increase? What happens when the child is distressed? What does the system remember? Who can change the rules? Which adult can intervene? What capability remains when the tool is gone?

Childhood cannot be delegated. The systems around it can be governed.

Chapter Eighteen carries this question into professional life. When AI supplies more of the answers, how do people develop and preserve the expertise needed to judge those answers?

Sources and Notes

1. Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı, and Rei Mariman, “Generative AI without guardrails can harm learning: Evidence from high school mathematics,” Proceedings of the National Academy of Sciences 122, no. 26 (2025): e2422633122, published June 25, 2025. https://www.pnas.org/doi/10.1073/pnas.2422633122. Accessed September 11, 2026. In a randomized field experiment with nearly 1,000 high-school mathematics students, standard GPT-style assistance improved performance while the tool was present but was followed by weaker unassisted exam performance; a teacher-designed guarded tutor largely removed the negative effect. The study is used here as the educational foundation for developmental delegation and the completion trap. Source record: DA-CH17-S1. Archive reference: pending.

2. Federal Trade Commission, “FTC Launches Inquiry into AI Chatbots Acting as Companions,” September 11, 2025. https://www.ftc.gov/news-events/news/press-releases/2025/09/ftc-launches-inquiry-ai-chatbots-acting-companions. Accessed September 11, 2026. The FTC used its Section 6(b) authority to order Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap, and xAI to provide information about companion-chatbot design, monetization, character development, testing, age restrictions, disclosures, data use, and potential negative effects on children and teens. The inquiry is a study and information-gathering process, not itself a finding that any recipient violated the law. Source record: DA-CH17-S2. Archive reference: pending.

3. California Senate Bill 243, Chapter 677, Statutes of 2025, Companion Chatbots, approved October 13, 2025; California Business and Professions Code Sections 22601-22606. https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB243. Accessed September 11, 2026. The law created California’s initial companion-chatbot framework, including disclosure, crisis-protocol, and minor-specific requirements, effective in 2026. Chapter Sixteen addresses the adult-facing provisions in more detail. Source record: DA-CH17-S3. Archive reference: pending.

4. Office of Governor Gavin Newsom, “Governor Newsom signs the strongest child safety chatbot and social media laws in the nation,” September 10, 2026; and California Senate Bill 1119, Companion chatbots: children’s safety, enrolled September 8, 2026. https://www.gov.ca.gov/2026/09/10/governor-newsom-signs-the-strongest-child-safety-chatbot-and-social-media-laws-in-the-nation/ and https://calmatters.digitaldemocracy.org/bills/ca_202520260sb1119. Accessed September 11, 2026. The governor announced signing SB 1119 as part of a child-safety package. The enrolled bill requires, beginning July 1, 2027, comprehensive child-safety risk assessments before new or substantially modified companion chatbots are made available, independent child-safety audits, specified enforcement, and a civil action for a child or parent in defined circumstances. The Governor’s release also records OpenAI’s public support for SB 1119. OpenAI systems are used in this Living Edition’s production workflow; that support is therefore presented as the company’s position, not independent validation. Source record: DA-CH17-S4. Archive reference: pending.

5. California Assembly Bill 1709, Chapter 183, Statutes of 2026, Covered platforms: age restriction: e-Safety Advisory Commission, signed September 10, 2026; related California age-assurance framework under AB 1043. https://legiscan.com/CA/text/AB1709/2025 and official California legislative materials linked there. Accessed September 11, 2026. The law prohibits covered platforms from providing defined addictive features, including addictive feeds and autoplay, to users under 16 and creates an e-Safety Advisory Commission. It operates with a coarse device-level age-bracket signal rather than requiring each covered service to collect identity documents directly and permits civil penalties of up to $50,000 per minor per knowing violation. The statute directs the commission to consider age-assurance privacy, differential impacts across youth, harmful design, safety practices, and safe access to online community for isolated LGBTQ+ youth. Source record: DA-CH17-S5. Archive reference: pending.

6. Character.AI, “Taking Bold Steps to Keep Teen Users Safe on Character.AI,” October 29, 2025, and “An Update On Changes to Our Under-18 Experience,” November 21, 2025. https://blog.character.ai/u18-chat-announcement/ and https://blog.character.ai/an-update-on-changes-to-our-under-18-experience/. Accessed September 11, 2026. Character.AI announced removal of open-ended chat for users under 18, following earlier teen filters, time-use notifications, Parental Insights, and age-assurance work. The company said the transition was informed by outside safety partners and described staged removal intended to reduce disruption. These are the company’s own descriptions of its product changes and motivations. Source record: DA-CH17-S6. Archive reference: pending.

7. OpenAI, “Introducing parental controls,” September 29, 2025, updated July 13, 2026; “Our approach to age prediction,” January 20, 2026, updated August 25, 2026; and “Introducing ChatGPT for Teens,” August 18, 2026, with linked under-18 model-behavior guidance. https://openai.com/index/introducing-parental-controls/, https://openai.com/index/our-approach-to-age-prediction/, and https://openai.com/index/chatgpt-for-teens/. Accessed September 11, 2026. OpenAI describes linked parental controls, quiet hours, safety notifications, age prediction, and a teen-specific ChatGPT experience with additional safeguards. Its under-18 rules bar romantic language, discourage emotional dependence, and prohibit the model from claiming feelings or consciousness. Parent controls do not provide access to a teen’s conversation transcripts. Adults misclassified as under 18 can verify age through a third-party identity-verification service. These are provider descriptions, not independent findings of effectiveness. OpenAI systems are also used in this Living Edition’s production workflow; see the transparency statement. Source record: DA-CH17-S7. Archive reference: pending.

8. Common Sense Media, “Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions,” July 16, 2025. https://www.commonsensemedia.org/research/talk-trust-and-trade-offs-how-and-why-teens-use-ai-companions. Accessed September 11, 2026. The national survey reports that 72% of teens had used AI companions at least once, 52% used them regularly, 33% had chosen an AI companion over a person for a serious conversation, and 24% had shared personal information. Common Sense Media is an advocacy organization that recommends against companion use by minors; the chapter therefore uses the survey primarily for prevalence and reported behavior and labels the organization’s normative position. Source record: DA-CH17-S8. Archive reference: pending.

9. UNICEF Office of Strategy and Evidence Innocenti, “Guidance on AI and children,” Version 3.0, current edition. https://www.unicef.org/innocenti/reports/policy-guidance-ai-children. Accessed September 11, 2026. UNICEF’s child-centered AI guidance calls for regulation and oversight, safety, privacy, fairness, transparency, child rights, developmental wellbeing, inclusion, skills, and an enabling environment. Version 3 adds attention to generative AI, AI companions, harmful data and content, and other emerging risks. The guidance is a policy framework grounded in child-rights principles, not a binding legal standard. Source record: DA-CH17-S9. Archive reference: pending.

10. Office of Governor Gavin Newsom, September 10, 2026 child-safety package announcement; Reuters, “California enacts new curbs on social media for children,” September 11, 2026. https://www.gov.ca.gov/2026/09/10/governor-newsom-signs-the-strongest-child-safety-chatbot-and-social-media-laws-in-the-nation/ and https://www.reuters.com/legal/litigation/california-enacts-new-curbs-social-media-children-2026-09-11/. Accessed September 11, 2026. The package included strengthened companion-chatbot safeguards, restrictions on addictive platform features for children, privacy and school-data measures, and restrictions involving AI-enabled toys. Reuters also reported criticism from digital-rights groups concerning speech and privacy. The chapter uses those criticisms to frame the tradeoff between protection, privacy, and access rather than to resolve the constitutional questions. Source record: DA-CH17-S10. Archive reference: pending.

11. Reuters, “British Columbia sues OpenAI over Tumbler Ridge school shooting,” September 21, 2026. https://www.reuters.com/legal/government/british-columbia-sues-openai-over-tumbler-ridge-school-shooting-2026-09-21/. Government of British Columbia, Ministry of Attorney General, “Attorney General’s statement on B.C. filing legal action against OpenAI,” September 21, 2026. https://news.gov.bc.ca/releases/2026AG0067-001105. Accessed September 22, 2026. The filing and its allegations are distinguished from judicial findings. Reuters reports OpenAI’s position that it works with government and law enforcement and continues safety work. The example concerns institutional detection, escalation and intervention; it does not reconstruct individual victims’ experiences or establish liability. OpenAI systems are used in this Living Edition’s production workflow; the company response is presented as its position, not independent proof of safety. Source record: DA-CH17-S11. Archive reference: pending.

12. Common Sense Media, Youth AI Safety Institute, “Tom Siegel joins as Executive Director of the Youth AI Safety Institute,” September 22, 2026. https://institute.commonsensemedia.org/tom-siegel-joins-as-executive-director-of-the-youth-ai-safety-institute. Institute homepage: https://institute.commonsensemedia.org/. Accessed September 23, 2026. The announcement confirms the appointment, a program to establish standards and open-source evaluations, independent product testing, and the crash-test-ratings analogy. The Institute launched in May; this is not a September launch or evidence that a complete binding standards regime now exists. The manuscript’s creation language describes standards development, and its car-seat comparison and institutional obligations are the author’s analogy and normative argument. The Institute’s advocacy and assessment claims are not proof of causal developmental harm or regulatory enforcement. Source record: DA-CH17-S12. Editorial brief item 2, September 23, 2026; formal Update 045. Archive reference: pending.

13. Internet Watch Foundation, “More AI-generated child sexual abuse images found in the first six months of 2026 than all of 2025, IWF warns,” October 5, 2026. https://www.iwf.org.uk/news-media/news/more-ai-generated-child-sexual-abuse-images-found-in-the-first-six-months-of-2026-than-all-of-2025-iwf-warns/ Accessed October 5, 2026. The assessment records material identified and assessed by IWF analysts, not the total amount online. Hash matching applies to previously identified images. The proposed framework for known and previously unseen material is IWF advocacy, not enacted law or proof of control effectiveness. Used at pattern level only, without details of individual minors or abuse methods. Source record: DA-CH17-S13. Update 068. Archive reference: pending.

14. Common Sense Media, Youth AI Safety Institute, “ChatGPT for Teens Risk Assessment,” updated October7,2026. https://institute.commonsensemedia.org/risk-assessments/chatgpt-teens. Barbara Ortutay, Associated Press, “Some guardrails on ChatGPT for Teens don’t work as promised, watchdog group says,” October7,2026. https://apnews.com/article/chatgpt-teens-safety-openai-7dba63edc37ee9764166c4115523daf0. OpenAI Help Center, “Managing parental controls in ChatGPT,” current guidance. https://help.openai.com/en/articles/12315553-managing-parental-controls-in-chatgpt. All accessed October8,2026. Account tests are not population harm rates. The Institute’s July–September testing does not evaluate the October GPT-6 release. The assessment includes an activation-window dispute; AP carries OpenAI’s response. The Institute is an advocacy organization receiving industry funding, including OpenAI funding, and its risk rating is not a legal finding. OpenAI’s current guidance limits safety notifications and acknowledges activation delay. OpenAI systems are used in this Living Edition’s production workflow; see the transparency statement. The parent-facing verification recommendation is author analysis. Prospective source record: DA-CH17-S14. Formal update number: 075. Archive reference: pending.