Circuit Breaker Labs builds 'crash-test dummies' to test AI for psychological harm

The startup Circuit Breaker Labs, one of TechCrunch’s 2026 Startup Battlefield 200 finalists, has built a testing platform it likens to an army of crash-test dummies for artificial intelligence. Its AI agents mimic users of different ages, backgrounds, languages and cultures to check whether models can spot psychologically dangerous conversations. The company will pitch at TechCrunch Disrupt in San Francisco on October 13-15.

Why the founders started it

Earlier this year, Character.AI settled several wrongful death lawsuits brought by families of underage users who died by suicide after interactions with its bots, and several families have sued OpenAI over ChatGPT’s alleged role in their loved ones’ suicides and delusions. The founders, siblings Shirali and Arul Nigam, say they were motivated by Sewell Setzer, a 14-year-old who became emotionally attached to a Character.AI chatbot and confided thoughts of self-harm to it before his death. In a 2024 lawsuit, his parents alleged that the bot encouraged him.

“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,” said Arul Nigam, the company’s CTO. He added that a bot may not have understood what words like “I want to be with you” really implied.

How the red-teaming works

Circuit Breaker Labs works with human domain experts to build hyper-realistic user simulations and runs adversarial red-team tests designed to reflect real speech patterns, slang, coded language and typos. Age, dialect and slang matter: a six-year-old girl, a 45-year-old man, someone who speaks English as a first versus a second language, or a gamer-slang user can each trip up a model, said CEO Shirali Nigam. “Models are really good at handling standard speech patterns, but nobody actually talks like that,” she said. The platform runs tens of thousands to hundreds of thousands of simulated interactions per day, and a proprietary scoring method turns the results into auditable, explainable scores.

What comes next

The startup currently acts as an AI safety testing lab for high-risk applications, including AI coaching, journaling and mental health support apps; Arul Nigam declined to name its marquee customers. It is still very early, with only five employees, including the two founders. Eventually, the team expects the platform to be used for any app where someone risks forming a parasocial relationship with a chatbot, including AI “co-worker” agents.

Arul Nigam said growing skepticism toward AI is healthy, but banning a potentially valuable tool over safety concerns would be “regressive.” “We want to help build that trust for people,” he said.