As AI leaders call for slowdown, children need a voice

The leaders competing to build the world’s most powerful AI systems agree that the technology may be moving too fast—and policymakers should pay attention. Anthropic CEO Dario Amodei has called for collective action to slow the pace of development, warning that safeguards are not keeping up. Elon Musk and OpenAI’s Sam Altman have chimed in with agreement.

The AI debate may sound far removed from the everyday lives of families with young children. It isn’t. AI is already showing up in children’s videos, learning tools, and toys—including social companion toys that can hold conversations, remember past interactions, and even respond in ways that can feel remarkably human. 

Altman has spoken personally about wanting his young son to “play in the dirt” and delaying his exposure to AI for as long as possible. His instinct is understandable. Young children build empathy, imagination, and social skills through responsive relationships with parents, caregivers, siblings, and peers. Technology can support those relationships, but it should not quietly displace them, especially when we still know so little about how repeated interactions with AI may affect children’s development. Current evidence indicates that only 5.26% of the current published research explores AI’s impacts on relationship formation and attachment. 

Many parents are asking versions of the same question as part of the Brookings Generation AI starts early initiative: What are the types of AI in young children’s lives? How much AI do I want in my young child’s life, and when? Parents should have the information and freedom to make thoughtful choices for their families. But they cannot be expected to assess rapidly changing technologies, manage widespread economic disruption, or make up for the absence of clear standards on their own.

Amodei’s proposal focuses on frontier AI development, not specifically on the toys, apps, and content entering young children’s lives. But the collective action problem he identifies extends to childhood: Even companies that want to proceed cautiously may find it difficult to slow down when their competitors do not.

As industry leaders call for a slowdown and lawmakers debate new AI safeguards, young children cannot be treated as an afterthought. Decisions are being made now, before we fully understand how these technologies affect children’s learning, relationships, and development. 

Existing protections are not enough to meet this moment. The Children’s Online Privacy Protection Act (COPPA) governs how companies collect data from children under 13, but says nothing about whether a product’s design is developmentally appropriate for a 3-year-old. Federal toy safety standards address choking hazards and lead paint, not a toy that converses, remembers, and responds like a friend. The Federal Trade Commission has opened an inquiry into AI companion chatbots, and some states have begun to act, but these efforts focus largely on teenagers—and proposals to preempt state AI laws could erase even this patchwork before a federal floor exists. Children from birth to 8 years, whose brains are developing faster than at any point in life, fall through the gaps. 

This moment offers an opportunity to ensure that parents, early childhood professionals, and child development experts are in the room to help shape how AI products that affect children are designed, tested, and governed. Their voices belong not only in forums focused specifically on children, but also in broader economic and policy discussions—from Senate Finance Committee hearings to the World Economic Forum—where decisions about how to share the wealth that AI generates and how to manage its disruption to work will shape the lives of children and families for years to come. 

Responsibility for navigating these changes cannot rest with parents alone, nor can we expect individual companies to solve the problem voluntarily. Even well-intentioned companies may hesitate to slow down or invest in stronger safeguards if they fear competitors will move faster. 

That is precisely why policymakers—in Congress and at agencies such as the Federal Trade Commission and the Consumer Product Safety Commission—have a role. Sensible guardrails can establish a shared floor, allowing companies to invest in safer, more developmentally appropriate designs without putting themselves at a competitive disadvantage. We have done this before; we don’t ask parents to crash-test their children’s car seats. We set federal safety standards, require testing before products reach the market, and let parents choose among options that have already met the floor. Automakers once resisted safety rules; today they compete on safety ratings. AI products built for young children deserve the same approach. We wouldn’t make tax policy without consulting businesses and economists. Why would we set policy that will fundamentally shape early childhood without parents and experts in the room? 

Policymakers may not be able to write a new rule for every technological development, but they can establish durable requirements for how AI products affecting young children are designed, tested, and reviewed in ways that are age-appropriate and evidence-based. Products likely to be used by children should undergo child-impact assessments before launch, examining not only privacy and content but how a product affects attachment, language, play and a young child’s ability to tell people from machines. Parents, early childhood professionals, and developmental experts should have formal roles in standards-setting, product testing, and regulatory review. 

Other countries are already moving: The United Kingdom requires online services likely to be accessed by children to assess the risks they pose, and the European Union has recently proposed the EU Kids Act. The United States has no federal equivalent. 

AI will also shape early childhood beyond the products that reach children’s and parents’ hands. Children live in families whose livelihoods may be transformed by AI and in communities affected by the environmental footprint of data centers. They rely on health care, housing, nutrition, early care and learning, and other systems being transformed by AI. We know children’s development is deeply affected by poverty, public health, and economic instability. Proposals to distribute the wealth generated by AI, update social insurance, address environmental health, or incorporate AI into systems meeting basic human needs should therefore be evaluated in part by how they affect young children. The participation of parents and experts should be required, not ceremonial.

A slowdown would give policymakers time to establish that shared floor and put the needs of babies and young children at the center of decisions about how AI is designed, tested, and governed. For young children, a slowdown need not mean a ban. It should mean that AI products designed for or marketed to children under 8—especially those built to act as companions—undergo independent developmental impact assessment before reaching the market, and that independent researchers can access the data needed to study them. Some AI tools may ultimately help young children, supporting language and learning, but children should not become the test cases for technology released before we understand its effects.

Leaving young children out of the AI debate will not shield childhood from AI. It will let market incentives, rather than developmental science, public interest, and informed parents, set the terms of children’s exposure. The question is no longer whether our children will encounter it, but what kind of AI childhood we are creating.