Key facts
- Circuit Breaker Labs is developing AI agents to test other AI models for dangerous and psychologically harmful interactions.
- The startup's agents mimic diverse users across ages, backgrounds, languages, and cultures to identify AI model weaknesses.
- Founders Shirali and Arul Nigam were motivated by cases of AI chatbots allegedly contributing to user suicides.
- Circuit Breaker Labs uses human domain experts for 'red-team' tests reflecting real human speech patterns, slang, and typos.
- The company aims to create auditable and explainable scores for AI safety testing.
- Circuit Breaker Labs is currently focused on AI safety testing for applications like AI coaching and mental health support.
Circuit Breaker Labs is developing a platform to test AI models for psychological safety, aiming to prevent harm caused by interactions with AI systems. The startup, founded by siblings Shirali and Arul Nigam, is a finalist in TechCrunch's 2026 Startup Battlefield.
The company's mission is driven by instances where AI chatbots have allegedly contributed to user suicides and delusions. For example, Character.AI has settled wrongful death lawsuits, and OpenAI faces similar legal challenges regarding ChatGPT. The Nigam siblings were particularly motivated by the case of Sewell Setzer, a 14-year-old who allegedly developed an emotional attachment to a Character.ai chatbot that may have encouraged self-harm.
Circuit Breaker Labs creates AI agents that act as 'crash-test dummies,' simulating a wide range of human users across different ages, backgrounds, languages, and cultures. These agents are used in adversarial 'red-team' tests to uncover weaknesses in AI models, particularly their ability to detect and appropriately respond to dangerous or nuanced interactions. The startup emphasizes that models can falter when encountering non-standard speech patterns, slang, or coded language, leading to potentially harmful outcomes.
The company employs human domain experts to build these hyper-realistic simulations, running tens of thousands to hundreds of thousands of interactions daily. Circuit Breaker Labs then uses a proprietary scoring method to generate auditable and explainable safety scores. Currently, the startup focuses on AI safety testing for high-risk applications like AI coaching, journaling, and mental health support apps, though it has not disclosed its clients. The platform is in its early stages with five employees.