Youth advocates argue in MIT Technology Review that state chatbot safety laws target catastrophes while ignoring everyday AI harms

• MIT study design. 54 adults aged 18–39 were split into three groups: LLM, search engine, and no tools. They wrote essays over three sessions while wearing EEG caps, and essays were analyzed with NLP and scored by humans and AI. In a fourth session, 18 participants switched conditions. Those moving from LLM to no tools showed reduced alpha/beta connectivity (arXiv).
• What SB 1119 requires of products. Child accounts must have persistent memory and push notifications off by default. Default limits are one hour per continuous session and two hours per day. Operators must either verify age through California's age-assurance framework or apply the child protections more broadly (Substack summary; leginfo).
• What other state laws require. All 12 laws require detecting self-harm signals and referring users to 988. Most also require recurring "you are talking to an AI" reminders, every one to three hours depending on the state (MultiState).

• "A third of teens" choose AI for serious matters. The survey measured whether teens had done this at least once; it did not measure a habit. It also found that 80% of teens say they put human friendships first (Common Sense Media; KUOW/NPR).
• "Cognitive debt" persists. This comes from a small preprint, and its lead author has said the study did not find "brain rot" and did not measure IQ (Tech & Learning).
• Google shaped legislation. Transformer's reporting supports this for Hawaii, Nebraska and Arizona. Google did not comment (Transformer).
• Hiner's coalition passed laws in eight states. This comes from the authors' bio. No independent corroboration was found, though Public Citizen did co-write a model bill with the Young People's Alliance (Public Citizen).

The op-ed applies arguments from the social-media era to AI: engagement-optimized design, and Meta's platforms being linked to violence in Myanmar, Sri Lanka and Ethiopia. In 2026, companion chatbots were the main AI issue in state legislatures. MultiState counted 12 states with enacted laws by June, with Hawaii's bill awaiting signature and Arizona's vetoed (MultiState). The authors are advocates. Leiken was previously TikTok's global head of youth safety and well-being policy (MIT Technology Review).

• Nebraska. The op-ed says the industry playbook "worked" there. Transformer reported in March that Nebraska's Google-backed bill looked likely to die in committee. LB 1185's provisions were later moved into LB 525, but the reviewed sources do not show whether that is the same bill (Transformer; AI Laws by State).
• "Low bars" vs. California. SB 1119's independent audits and use limits go beyond the self-harm and sexual-content minimum the op-ed criticizes (Office of the Governor).
• The evidence on cognition and teen behavior is early. The MIT study had 18 participants in its fourth session, and the teen survey measured self-reported, at-least-once behavior.

• Compliance now varies by state. Builders face different reminder schedules, age-assurance rules, crisis protocols and attorney-general enforcement in each state. Nebraska's penalties start at $1,000 per violation, up to $500,000 per operator (AI Laws by State).
• California sets the high bar. Defaults that limit engagement and third-party audits under SB 1119 are where the standard is heading, and the advocates explicitly want more friction in products.
• Exemptions may not last. Assistants built into other products, and customer-service or productivity tools, are excluded from many state definitions for now (MultiState). Advocates are pushing to close those gaps.
• EdTech should expect evidence demands. Schools are being asked to require evidence that AI improves learning before adopting it widely.