Table of Contents
Published: April 3, 2026
Read Time: 4.4 Mins
Total Views: 230
In the public conversation about artificial intelligence and human welfare, the discussion tends to gravitate toward long-term existential risks — the scenarios involving catastrophic or extinction-level events that dominate both the most alarming headlines and the most serious academic debates. Those risks deserve the attention they receive, and I have written about them in the context of biosecurity. At the same time, a more immediate hazard is accumulating right now, at a scale that is already measurable in lawsuits, legislative responses, and documented harm: the impact of AI companion chatbots on mental health, and particularly on the mental health of adolescents and young adults.
Millions of people now interact daily with AI systems designed to simulate friendship, empathy, or therapeutic dialogue. These systems are engineered specifically for emotional realism — for the capacity to generate responses that feel like understanding, connection, and care. For some users, particularly those experiencing isolation or psychological distress, that emotional realism creates genuine attachment and dependence. Some of those users are minors. Some of them disclose suicidal thoughts to these systems. In multiple lawsuits, plaintiffs have alleged that interactions with AI chatbots contributed directly to users taking their own lives.
California’s Legislative Response and Its Limits
In 2025, California enacted Senate Bill 243, becoming the first state in the nation to regulate AI companion chatbots. The law requires clear disclosure when a user is interacting with AI rather than a human, mandates protocols for responding to suicidal ideation, imposes additional protections for known minors — including periodic reminders that the chatbot is not human and restrictions on sexually explicit content — and creates a private right of action allowing civil lawsuits for violations. As a first legislative effort to address this specific category of harm, the law represents a meaningful acknowledgment that emotionally realistic AI systems require regulatory oversight tailored to their particular risks.
The honest assessment, though, is that the effectiveness of these measures remains genuinely uncertain. Disclosure that a system is not human may be insufficient to counteract the emotional realism that makes these systems compelling to vulnerable users in the first place; a person in crisis who experiences a chatbot as empathetic and responsive may not be protected by a disclosure they read and accepted during account creation. Protocol requirements for responding to suicidal ideation do not guarantee that those protocols will be effective when implemented by systems that were designed primarily for engagement rather than clinical safety. Operators may structure their data collection practices to minimize what they know about users’ ages, thereby limiting the obligations they trigger under the law’s provisions for minors.
The Social Media Precedent and What It Predicts
From a public health perspective, the trajectory of this issue tracks uncomfortably closely to the history of social media regulation, and that parallel is instructive rather than reassuring. Social media platforms optimized for engagement have amplified loneliness, misinformation, and depression at a population scale that researchers are still working to fully characterize. Nearly sixteen years after Instagram was released, Meta CEO Mark Zuckerberg was required to testify in court about whether the platform was deliberately designed to addict and harm teenagers. The legal accountability that advocates spent years pursuing arrived long after the harms had accumulated across an entire generation of young people.
AI companion systems have the potential to replicate that pattern at an accelerated pace and broader scale. The engagement optimization that characterized social media’s most harmful dynamics is, in many ways, structurally built into the design of companion chatbots — systems whose commercial success depends on users feeling sufficiently connected to return repeatedly and to maintain long-term relationships with AI entities. The mental health risks that flow from that dynamic, particularly for adolescents whose social and emotional development is still in progress, are not speculative. They are the same risks that social media researchers documented across the 2010s, now potentially amplified by a technology that is more conversational, more personalized, and more emotionally convincing than any social media feed.
Public health agencies have always been required to operate on two timescales simultaneously: preparing for rare catastrophic events whose consequences would be irreversible, while addressing the everyday harms that are accumulating in real time. The AI era does not change that fundamental structure. It simply adds new categories of harm to both ends of the spectrum and demands that the institutions responsible for protecting population health develop the technical literacy, the regulatory frameworks, and the research infrastructure to address them before another generation bears the cost of delayed action.
