Circuit Breaker Labs hopes to make AI safer for your kids (and you)

Circuit Breaker Labs builds simulated user agents it likens to an "army of crash-test dummies" to evaluate conversational AI for psychologically risky interactions. The company says its agents represent a wide range of ages, backgrounds, languages and cultural speech patterns so models can be tested against realistic, everyday ways people actually communicate, including slang, coded language and typographical errors. The startup assembles these hyper-realistic simulations with the help of human domain experts and uses them to run adversarial "red-team" tests. Circuit Breaker Labs reports running tens of thousands to hundreds of thousands of simulated interactions per day, then applying a proprietary scoring system to generate auditable, explainable safety scores that track how models respond to dangerous or ambiguous situations. Founders Shirali and Arul Nigam framed the company’s mission around real-world harms, pointing to cases where underage users formed emotional attachments to chatbots and later harmed themselves. The team emphasizes that models often handle standard speech patterns well but can fail when faced with nuance, slang or language differences — failures that can have serious psychological consequences for users seeking support from AI. Currently operating as an AI safety testing lab, Circuit Breaker Labs focuses on high-risk applications such as AI coaching, journaling and other mental-health–adjacent tools. The company has a working product but is in an early stage with five employees; the founders declined to name marquee customers. They envision the platform eventually testing a broader array of AI agents, including workplace or co‑worker bots, where repeated interactions can foster parasocial relationships and unpredictable behavior.