Congress Is Looking at the Wrong AI Risk

 As technology advances, AI Chatbots are increasingly falling into the hands of younger generations. Nearly nine in ten kids ages 9 to 17 use or interact with AI, with roughly one in four doing so daily. 

Legislation has been proposed to implement guardrails to protect children from potential harms and abuses from this rapidly evolving technology – but the most recent efforts still aren’t explicitly regulating what we should be protecting kids from. 

For instance, last month, the House passed the KIDS Act, a bill meant to protect children and teens online. AI constitutes a portion of the KIDS Act, but the major aspects of the legislation are directed at social media platforms and not at AI-specific risks. 

The KIDS Act appears to be the culmination of legislative efforts over the course of the past several months. Provisions in the Act were drawn from the SAFE Bots Act – which requires a disclosure that a platform is AI-operated and a break prompt after three hours of use of an AI chatbot – and the AWARE Act – which creates AI literacy resources for parents, teachers, and children. 

A glaring omission of the KIDS Act, however, is its failure to address an important child safety issue: sycophancy. 

Sycophancy is defined by the excessively agreeable, affirming quality built into generative AI chatbots. Children are turning to AI chatbots during some of the most formative moments of their emotional and social development, and what they receive in return is consistent, unwavering affirmation. 

In 2025, companionship and therapy were the most popular uses of generative AI. Thirty percent of teens reported talking to AI instead of a real person for serious conversations. The use of generative AI is only expected to continue to increase, and so too will the number of children who are able to access and use it for emotional support. 

When presented with interpersonal conflicts, sycophantic AI models affirm users’ actions approximately 50% more than humans do. Adult users of sycophantic AI reported higher perceptions of their own rightness as well as lower willingness to engage in behavior that would solve a conflict compared to users of non-sycophantic AI models. Participants additionally considered sycophantic responses as higher quality and more trustworthy – corresponding to an increased willingness to use the AI platform again. The feature that distorts judgement is also driving continuous engagement. 

If adults – who have had years of experience developing critical thinking skills – are affected by sycophantic AI behavior, the implications for children can only be more serious. 

Children’s brains respond to rewards more reactively than developed adult brains. This makes reward-seeking mechanisms especially sensitive for children and teens, resulting in more impulsive, reward-chasing behavior. 

The affirmation sycophantic AI models provide serves as a reward for children. They are receiving this stimulation at a point in their emotional development when their capacity to critically evaluate AI-generated content has not matured. 

AI is constantly accessible, and users are comforted by the fact that AI platforms are affirming – by design, generative AI can’t judge or criticize someone for making a mistake. This leads people to share even more information with an AI platform than they would with another human. 

Children are adversely affected by this trust paradigm. In conversations about mental health, children were more willing to share sensitive or highly personal information with AI chatbots than they were with humans. 

An over reliance on AI as children could have lasting implications for their emotional development as an adult. A child who repeatedly uses AI models to discuss emotionally heavy topics and interpersonal conflicts may develop a stronger tendency to believe they are consistently acting correctly and in turn, a lesser willingness to take initiative to repair a relationship. 

Congress has recognized the existence of the sycophancy problem in the KIDS Act and in other pieces of proposed legislation. The GUARD Act, introduced in October 2025, for example, acknowledges that AI chatbots “can manipulate emotions and influence behavior in ways that exploit the developmental vulnerabilities of minors.” The CHATBOT Act, introduced in March 2026 would go further, mandating a study on “the prevalence and effects of sycophantic or excessively affirming behavior by artificial intelligence chatbots on children and teens.” 

The Youth AI Privacy Act, introduced in March 2026, comes closer, banning specific engagement mechanics like push notifications and streak badges for minor users. These are prohibitions on design features that encourage compulsive use, pulling a child back to the platform – but the conversational content remains untouched. Removing a streak badge does not change the fact that the bot will still consistently tell a child they were right, regardless of the circumstances presented to it. 

Provisions of the SAFE Bots Act require a break prompt after three continuous hours of use by a minor. Three hours of sycophantic interaction with a developing brain is already a long time, and a disclosure that the platform is AI-operated does not address what the AI is saying. Regardless of whether the user knows the AI platform is non-human, people will continue to engage with it as if it were another human. 

The KIDS Act demonstrates that this is not a question of legislative capacity. The Act requires that covered social media platforms exercise “reasonable care in the creation and implementation of any design feature” to prevent compulsive usage and related mental health harms – a standard regulating how the product is built. But this standard needs to be applied to the design of AI chatbots as well. 

We cannot let children communicate with sycophantic AI platforms more than they do with their parents or family members. A lack of regulation for these chatbots will foster a generation without the ability to critically think about their own actions and their relationships with other people. To protect children in this country, Congress needs to regulate the core product experience of every mainstream AI assistant and companion. 

Sarah Shahed is a Domestic Policy Programs intern for Summer 2026