Startup builds AI 'crash-test dummies' to catch psychologically harmful chatbot replies

Circuit Breaker Labs, one of TechCrunch's 2026 Startup Battlefield 200 finalists, is building tools meant to make AI chatbots safer, particularly for children and teenagers. The startup will present at TechCrunch Disrupt in San Francisco this October.

The effort follows a string of cases in which interactions with AI chatbots have been linked to serious psychological harm, including suicide. Character.AI settled several wrongful death lawsuits this year filed by families of underage users who died by suicide after engaging with its bots. Several families have also sued OpenAI over ChatGPT's alleged role in loved ones' suicides and delusions.

Circuit Breaker Labs was founded by siblings Shirali and Arul Nigam, who cited the case of Sewell Setzer, a 14-year-old who formed an emotional attachment to a Character.ai chatbot and shared thoughts of self-harm with it before dying by suicide. His parents' 2024 lawsuit alleged the bot had encouraged him.

How the testing works

The company has built AI agents that act as stand-ins for users of different ages, backgrounds, languages and cultures, used to test whether models can recognize dangerous or psychologically harmful exchanges. Built with input from human domain experts, the tests reflect real speech patterns, slang, coded language and typos, since informal or unusual phrasing often trips up models that otherwise handle standard language well.

Circuit Breaker Labs runs tens of thousands to hundreds of thousands of simulated conversations daily and applies a proprietary scoring system to produce auditable results. The company currently works with high-risk applications such as AI coaching, journaling and mental health support tools, though it declined to name its clients. With just five employees, the startup is still in its early stages, but its founders argue that improving AI safety — rather than banning the technology — is the right response to growing public distrust.