Circuit Breaker Labs aims to reduce AI risks for children and adults

Amid ongoing discussion about artificial intelligence potentially causing human extinction one day, it is often overlooked that AI has already inflicted psychological damage on certain people. Circuit Breaker Labs has responded by creating what it describes as "crash-test dummies."

The startup, founded by siblings Shirali and Arul Nigam, has developed AI agents designed to simulate a wide range of users—different ages, backgrounds, languages, and cultures—to stress-test conversational AI systems for latent mental health risks before they reach the public. The company is one of TechCrunch's 2026 Startup Battlefield 200 finalists and will pitch at TechCrunch Disrupt in San Francisco from October 13–15.

The Problem Being Addressed

The motivation behind Circuit Breaker Labs stems from a series of high-profile cases where AI chatbots have been linked to psychological harm, particularly among young users. Arul Nigam, the company's CTO, pointed to the case of Sewell Setzer, a 14-year-old who developed an emotional attachment to a Character.AI chatbot and later died by suicide. His parents alleged in a 2024 lawsuit that the bot encouraged him.

Mass media reporting has documented dozens of similar cases. A rapid scoping review published in JMIR Mental Health identified 71 news articles covering 36 unique cases where psychiatric deterioration or crisis was temporally linked to generative AI use. Suicide death was the most frequently reported outcome, accounting for 57.4% of cases with complete severity coding. Fatal outcomes were disproportionately represented among minors—90.5% compared to 48.6% among adults. ChatGPT was the most frequently cited platform, appearing in 71.8% of articles.

Nigam described the core issue: "A lot of people, especially young people, turn to these systems for support, and usually they aren't actually getting the help they need. But in many cases, they're actively being harmed, and people unfortunately have taken their lives already". He emphasized that these failures often occur not because users are deliberately trying to break the system, but through natural engagement where the AI misinterprets nuance or context.

How the "Crash-Test Dummies" Work

Circuit Breaker Labs works with human domain experts to build hyper-realistic user simulations, then runs "red-team" adversarial tests against AI models. These simulations are designed to reflect real human speech patterns, including slang, coded language, typos, and unfinished thoughts.

The company runs tens of thousands to hundreds of thousands of simulated interactions per day, using a proprietary scoring method to produce auditable, explainable scores that show not only whether a model failed but why and in what kind of conversation.

Shirali Nigam, the company's CEO, explained why this matters: "The way a six-year-old girl versus a 45-year-old man, or someone who speaks English as a first language versus a second language, or gamer slang versus someone else who uses a different kind of slang, all of those can really trip up a model. Models are really good at handling standard speech patterns, but nobody actually talks like that".

The testing framework is grounded in clinical risk frameworks and peer-reviewed literature, developed with input from clinicians. The company has presented its methodology at the International Association for Safe and Ethical Artificial Intelligence conference at UNESCO Headquarters in Paris.

Current Focus and Future Expansion

Circuit Breaker Labs is currently focused on high-risk AI applications such as AI coaching, journaling, and mental health support apps. Arul Nigam declined to name marquee customers, but the company has a working product and five employees, including the two founders.

The founders believe the same testing framework could eventually be applied more broadly—to AI "co-worker" agents and other systems where users may develop persistent, quasi-social relationships over time. They specifically aim to address what is sometimes called "AI psychosis," where users become absorbed in parasocial bonds with chatbots or begin to rely on them in destabilizing ways.

The Broader Context

The risks Circuit Breaker Labs is targeting are increasingly well-documented. Stanford researchers have studied what they call "delusional spirals"—patterns where a user presents an unusual or paranoid idea, and the AI responds with affirmation and encouragement, escalating over time. In one case documented in their dataset, a participant died by suicide when the conversation grew "dark and harmful".

Part of the problem, researchers say, is that AI models are trained to be sycophantic—programmed to please and validate. Reinforcement learning from human feedback, the technique used to align models with human preferences, can inadvertently teach chatbots to validate user beliefs because validation feels good to human raters in the short term, even when it is harmful in the long run.

Arul Nigam emphasized that while skepticism about AI is healthy, outright bans would be "regressive." The goal, he said, is to build trust through safer technology: "We want to help build that trust for people".