Ritter Calls for AI Chatbot Safety Standards

Harry Ritter, Founder and CEO of Alma and Chief Strategy Officer at Spring Health, argues that AI chatbots used for mental health support should face independent safety checks and clear accountability. He says clinicians are held to a duty to avoid harm, while companies building chatbots currently decide for themselves whether their systems are safe, with no outside review beyond voluntary pledges.

A RAND study published in JAMA Pediatrics found that nearly one in five young people ages 12 to 21 have used AI chatbots for mental health advice, up from one in eight one year earlier. Data released by OpenAI indicate that 0.15% of ChatGPT users, more than 1.35 million people, express suicidal intent or plans to a chatbot each week. Ritter says chatbot responses in crises vary widely and there is no universal way to measure their safety.

Ritter compares voluntary chatbot oversight with social media companies’ self-regulation after researchers warned about effects on young people’s mental health a decade ago. He says meaningful action came only after years had passed and congressional oversight arrived too late, with damage and deaths already occurring. He argues that safety should be demonstrated before products are sold, as in pharmaceuticals, food safety and aviation.

California, New York, Oregon and Hawaii have passed chatbot safety laws, Ritter says. The laws require chatbots to follow protocols when users express suicidal thoughts and refer them to crisis resources, including for large language models where he says most dangerous conversations take place. He says differing state laws cannot protect every teenager in crisis and calls for a federal standard for safety and liability, supported by independent evaluations developed by experts. Spring Health launched VERA-MH, an open-source evaluation standard developed with academic and industry experts, as a basis for objective third-party safety and transparency tools.

Ritter also proposes requiring platforms to help users in crisis reach human support through trusted provider networks. He says platforms that meet safeguards should receive liability protections unavailable to those that do not. He frames the proposal as accountability for how AI is used, arguing that companies should prove safety when people use general-purpose models for mental health support, even though those models were not designed for that purpose. Young people may turn to chatbots because they are free and accessible when real mental health care is unavailable, he says; systems should meet a basic safety standard and help them or connect them with a human.

Harry Ritter, Founder and CEO of Alma and Chief Strategy Officer at Spring Health, argues that AI chatbots used for mental health support should face independent safety checks and clear accountability. He says clinicians are held to a duty to avoid harm, while companies building chatbots currently decide for themselves whether their systems are safe, with no outside review beyond voluntary pledges.

A RAND study published in JAMA Pediatrics found that nearly one in five young people ages 12 to 21 have used AI chatbots for mental health advice, up from one in eight one year earlier. Data released by OpenAI indicate that 0.15% of ChatGPT users, more than 1.35 million people, express suicidal intent or plans to a chatbot each week. Ritter says chatbot responses in crises vary widely and there is no universal way to measure their safety.

Ritter compares voluntary chatbot oversight with social media companies’ self-regulation after researchers warned about effects on young people’s mental health a decade ago. He says meaningful action came only after years had passed and congressional oversight arrived too late, with damage and deaths already occurring. He argues that safety should be demonstrated before products are sold, as in pharmaceuticals, food safety and aviation.

California, New York, Oregon and Hawaii have passed chatbot safety laws, Ritter says. The laws require chatbots to follow protocols when users express suicidal thoughts and refer them to crisis resources, including for large language models where he says most dangerous conversations take place. He says differing state laws cannot protect every teenager in crisis and calls for a federal standard for safety and liability, supported by independent evaluations developed by experts. Spring Health launched VERA-MH, an open-source evaluation standard developed with academic and industry experts, as a basis for objective third-party safety and transparency tools.

Ritter also proposes requiring platforms to help users in crisis reach human support through trusted provider networks. He says platforms that meet safeguards should receive liability protections unavailable to those that do not. He frames the proposal as accountability for how AI is used, arguing that companies should prove safety when people use general-purpose models for mental health support, even though those models were not designed for that purpose. Young people may turn to chatbots because they are free and accessible when real mental health care is unavailable, he says; systems should meet a basic safety standard and help them or connect them with a human.