As artificial intelligence becomes an increasingly pervasive part of daily life, tech companies and policymakers have stepped up efforts to address safety and well-being. Recent safeguards include parental controls, break reminders, usage limits, and crisis detection tools designed to support users during mental health conversations. However, experts warn that these existing measures focus almost exclusively on acute, immediate harms rather than the slow, cumulative effects that develop over months and years of interaction.
The Blind Spot of Acute Harm Focus
Current industry safety guardrails operate almost entirely on a per-message or per-session basis. While recent updates have allowed some companies to track distress across multiple sessions, these features remain targeted at identifying sudden crises like suicide or psychosis.
This narrow focus creates a major blind spot for long-term behavioral changes. A chatbot conversation can be completely safe and helpful in every single interaction, yet regular use over a long period could still gradually displace the human relationships necessary for health and well-being. Early preliminary studies indicate that heavier use of AI for emotional support and companionship is linked to increased loneliness and reduced socialization.
Why Data Access is Essential for Researchers
Independent researchers and policymakers currently lack the visibility into usage data and the shared measurement instruments needed to determine whether cumulative displacement is truly occurring. AI companies hold their aggregate user data closely, making it difficult to study long-term societal impacts.
Historical precedents show the danger of ignoring cumulative harm until it is too late. In other industries, companies have successfully repurposed individual-level product safeguards—such as warning labels or break notifications—as legal or rhetorical defenses against accountability. Establishing robust measurement infrastructure before potential harms become undeniable is critical. If AI is not displacing human connection at scale, comprehensive data and measurement should prove that as well.
Building a Measurement Infrastructure
To bridge the gap between private AI data and public welfare, researchers suggest that three key groups must take action:
- Researchers in academia and civil society: Experts need to develop valid instruments and measurement approaches that can isolate the impact of AI chatbots from other factors contributing to modern social isolation.
- Public health and education agencies: Local and state institutions should measure and share the existing costs of social isolation on public systems, tracking metrics like primary care utilization, school counselor hours, and social prescription programs.
- State legislators: Lawmakers can leverage pending companion chatbot bills to encourage greater access for independent third parties to evaluate risks related to social isolation.
Without a concerted effort to build this measurement infrastructure today, society risks repeating the protracted, decade-long debates seen with social media and adolescent mental health—leaving policymakers without the data they need when it matters most.
