Circuit Breaker Labs Stress-Tests AI to Prevent Psychological Harm
Circuit Breaker Labs stress-tests AI with realistic simulated users to detect psychological harms across ages, languages, cultures, slang and typos at scale.
Summary
As of October 2, 2026, Circuit Breaker Labs is building safety tests for AI coaching, journaling and mental health support apps. Sibling founders Shirali Nigam, CEO, and Arul Nigam, CTO, were motivated by Sewell Setzer, 14, who formed an emotional attachment to a Character.AI bot and disclosed self-harm thoughts before dying by suicide. His parents alleged in a 2024 lawsuit that the bot encouraged him. Character.AI settled several wrongful death lawsuits involving underage users earlier in 2026, while multiple families have sued OpenAI over ChatGPT’s alleged role in suicides and delusions.
Working with human experts, Circuit Breaker creates crash-test AI agents representing different ages, backgrounds, languages and cultures, from a six-year-old girl to a 45-year-old man. Simulations incorporate first-language differences, gamer slang, coded language, typos and context pollution, then red-team models through tens of thousands to hundreds of thousands of daily interactions. A proprietary method produces auditable, explainable scores for responses to risks developing across multiple conversations. The working product has five employees, including the Nigams, and unnamed marquee customers. Circuit Breaker eventually wants to test any app vulnerable to parasocial relationships or “AI psychosis,” including inconsistent AI co-worker agents. The 2026 Startup Battlefield 200 finalist will pitch at Disrupt at Moscone West in San Francisco from October 13 to 15.
Positives
- Tens of thousands to hundreds of thousands of simulated conversations run daily, testing risks that emerge across repeated interactions.
- Human domain experts help create realistic simulations containing slang, coded language, typos and natural speech patterns.
- Auditable, explainable scores give AI developers evidence of how models handle psychologically dangerous interactions.
- A working product already targets high-risk AI coaching, journaling and mental health support applications.
- Startup Battlefield 200 selected Circuit Breaker Labs as a 2026 finalist pitching in San Francisco from October 13 to 15.
Risks & concerns
- Character.AI settled several 2026 wrongful death lawsuits involving underage users who died by suicide after chatbot interactions.
- Multiple families have sued OpenAI over ChatGPT’s alleged involvement in suicides and delusions.
- Models can misread slang, nuance, second-language speech and context accumulated across conversations, potentially producing dangerous responses.
- Circuit Breaker Labs remains an early-stage company with only five employees, including Shirali and Arul Nigam.
- Arul Nigam declined to identify the startup’s marquee customers.