The AI for Good wrapped on Friday. I took my time to reflect because I felt uneasy that I may have been too blunt in my positions... But over the weekend, that unease resolved and I remembered why clarity matters.
Clarity isn't adversarial. It's the precondition for action, for collaboration, for effective policy-making. You cannot draw a line you haven't named. You cannot prevent harm you haven't defined. And when it comes to children and AI, being endlessly balanced in our positions is itself a choice, and not always a helpful one.
So here is where I am:
1. Children should not interact with anthropomorphic, general-purpose AI systems designed to mimic human relationships, empathy, attachment, closeness, and companionship. The developing brain is not a testing ground. António Guterres said it at the launch of the UN's Global Dialogue on AI Governance, and the UN Scientific Panel on AI linked sycophantic AI to documented mental health harm. The evidence is here. We must act.
2. This doesn't mean there are no beneficial uses of AI for children in health, education, environment. There are and they must be designed backwards from developmental outcomes, specific in purpose, not intended to create attachment. Drawing this line isn't anti-industry, it gives companies a clear signal on where societies want AI applied for children, and that enables responsible innovation & collaboration.
3. The burden of proof needs to move. Industry must demonstrate before and after deployment that what they're building is developmentally appropriate and beneficial. We have done this in every other domain where children's safety was at stake.
4. The conversation we need is upstream of where most safety work sits. Benchmarking what AI models generate at the prompt level matters but it's being misread as a signal that models are safe enough to deploy in schools, at homes. Without the contextual clarity, this work may end up producing unintended negative consequences.
Thank you to Fanny Rotino and the Child and Youth Friendly Governance Project for inviting me to moderate "Designing AI for Teenagers" with young experts Sarah Shafiq, MHD Nour Riz, Lorenzo Impellizzeri & Iren Neeranzona Khan, and professionals Henrietta Ridley, Wafa Ben-Hassine, Leanda Barrington-Leach, Prof. Salma Abbasi, and Zachary Stein. You can see the recommendations from 1000+ kids across the world on the photo.
Thanks to Human Change Foundation (Margarita Louis-Drey, Ileana Konialidis, Maria Veiga Barracosa Jenna G.) for convening a policy lunch attended by the President of Iceland H.E. Halla Tomasdottir and experts across sectors. Zachary Stein gave a powerful keynote on the dangers of attachment economy. Ron Ivey and I outlined a path forward. Tristan Harris closed with clarity on what's possible if we act now.
Every parent and caregiver knows this: children flourish within clear, supportive boundaries, not despite them. We are simply asking societies to do the same with AI.
