Circuit Breaker Labs Builds AI Safety Tests After 14-Year-Old's Chatbot-Linked Suicide
Updated
Updated · TechCrunch · Oct 2
Circuit Breaker Labs Builds AI Safety Tests After 14-Year-Old's Chatbot-Linked Suicide
3 articles · Updated · TechCrunch · Oct 2
Summary
Circuit Breaker Labs has built AI agents that simulate users across ages, languages and cultures to probe models for psychologically dangerous responses in high-risk apps such as AI coaching and journaling.
Tens of thousands to hundreds of thousands of simulated conversations run each day, using slang, coded language and typos to expose failures that standard testing can miss over long interactions.
The startup says it then assigns auditable, explainable safety scores, aiming to catch risks such as parasocial attachment and "AI psychosis" before products reach vulnerable users.
The five-employee company was motivated by cases including 14-year-old Sewell Setzer's death and follows lawsuits alleging Character.AI and ChatGPT interactions contributed to suicides, delusions and other harm.