“Is AI safe for kids?” is where the conversation starts. It can’t be where it ends.
Last week our team spent the day at The New York Public Library for the inaugural Common Sense Media Kids & AI Summit. We joined sessions throughout the day, but much of our time was spent sitting down with customers and EdTech teams, working through how their products actually handle student data.
The thread running through the day was a shift in posture. For years, the default response to kids’ privacy and safety has been fear, followed by a checklist. What we heard on Wednesday was a call for something more demanding: safety that is designed in, enforceable, and visible to the families and educators who rely on it.
The clearest message of the day was that parents and regulators are, for once, largely aligned. Neither group is willing to let AI become social media 2.0, with young people serving as the testing ground for products whose effects are only understood years later.
That alignment changes the timeline. With social media, protections arrived long after habits, business models, and harms were already set. With AI, families, educators, and lawmakers are asking hard questions while products are still being designed. For product teams, that means the choices made in the next few release cycles are likely to define how their tools are judged for years to come.
That is close to home for us. Evaluating whether a product earns trust, rather than simply asserting it, is the work we do every day. Here’s what stood out most.
A New Law Sets the Tone
The day opened with news. Governor Kathy Hochul joined the summit shortly after signing New York’s new law on AI chatbots and minors. Common Sense Media worked with the bill’s sponsors and the Attorney General’s office on its development.
The law targets specific design choices rather than chatbots as a category. Developers may not offer minors features that impersonate a human, use sycophantic responses that build emotional dependency, draw on a young person’s past health or mental health data, or encourage self-harm, violence, or keeping secrets from trusted adults.
For those of us who spend our days reading privacy policies and terms of service, a few details stood out:
- Data is part of the harm model. Using a child’s past health or mental health information to shape a chatbot’s responses is named directly.
- Age checks come with data limits. Companies must confirm a user is an adult before turning on restricted features, and delete age-verification information after use.
- Consequences reach the data itself. Reported penalties run up to $25,000 per violation, with possible orders to destroy unlawfully collected data or AI models trained on it.
- The scope is broad. Analysis of the bill notes it reaches general-purpose assistants that hold ongoing conversations, not only companion apps.
Whatever one’s view of the policy, the signal for product teams is clear. Privacy and safety are increasingly evaluated together, at the level of individual features.
Asking a Better Question
Speakers including former U.S. Surgeon General Dr. Vivek Murthy, along with youth and policy advocates, kept returning to one idea. “Is AI safe for kids?” is the wrong place to stop. The better questions are who teaches young people to use these tools well, and how companies build safety in by default.
We heard that framing echoed in our working sessions. When we sat down with EdTech teams to look at their privacy posture, the most useful conversations moved past “Are we meeting the minimum requirements” to more specific questions:
- What student data does this feature actually need?
- What happens to that data when a school ends its contract?
- Is the protective setting the default, or something a teacher has to find?
- Would a parent understand their choices without a lawyer?
The teams that had strong answers had usually made deliberate design choices early, like collecting less, keeping retention short, and separating the student experience from everything else. Those choices protect students first. They also make a product easier to explain and easier to trust.
Why “Trust Us” Isn’t Enough
Several sessions looked hard at the limits of industry self-regulation. Panelists pointed to recent testing that found gaps in teen modes on major AI platforms, and to lawsuits alleging that “reset” buttons on social media did less than users were led to believe.
The lesson the room drew was not that every company acts in bad faith. It was that a protective feature is only as good as the evidence behind it. A setting that exists on paper but fails in practice gives families false confidence, which can be worse than no setting at all.
That gap between what a product says and what it does is exactly where independent evaluation matters. When we review a product, we evaluate the privacy documents against a consistent rubric, and we look for specifics: what is collected, who it is shared with, how long it is kept, and what control families and schools actually have. Law sets a floor. Independent review helps families and educators see how far above that floor a product really is.
A quieter theme surfaced in conversations between sessions: real skepticism about whether AI belongs in every product it is being added to. That skepticism is not simple resistance to new technology. It comes from a lack of trust, a lack of understanding about how these tools work, and a sense that AI is often being implemented without clear intention.
For EdTech teams, that has a practical consequence. Adding “AI” to a product tagline is not enough, and for many parents and educators it may now raise more questions than it answers. A label invites the reader to wonder what data the feature uses, what it is for, and whether anyone checked that it works.
What carries weight instead is evidence of two things:
- Trust. Clear answers about what data an AI feature uses, how long it is kept, and what control families and schools have.
- Impact. A specific account of the problem the feature solves for students or teachers, and how the team knows it helps.
Explaining why a feature exists, not just that it exists, is a higher bar than a tagline. It is also the bar families and schools are increasingly asking products to clear.
Keeping People at the Center
Some of the most memorable moments came from students. Alongside educators and researchers, student panelists talked about protecting real relationships, mental well-being, and the ability to think for themselves. They did not want AI to become an emotional crutch or a stand-in for growing up alongside other people.
That perspective is a useful check on privacy work too. Data minimization and clear retention limits are not abstractions. They shape what a tool knows about a young person, and how much a child is encouraged to share with a product rather than with the adults in their life.
What Product Teams Can Do Next
We left the library with a fuller notebook and a clearer sense of direction. The conversation about kids and technology is moving from reaction to design, and it is moving while AI products are still taking shape. More states are writing specific expectations into law, and families and schools are asking sharper questions about the products in their classrooms.
For EdTech companies, that is an opportunity. Teams that have already made careful choices about student data, and can explain the purpose behind their AI features, are well placed to show that work plainly, with evidence. We will keep doing our part by reviewing products independently and sharing what we find in language families can use.
Thank you to Common Sense Media for convening the summit, and to every customer and partner who sat down with us to dig into the details.

