Circuit Breaker Labs hopes to make AI safer for your kids (and you)
TechCrunch Julie Bort
Circuit Breaker Labs is building tests to catch AI that turns harmful, especially for kids. It simulates real people and languages, because bad nuance can get dangerous fast.
Based on reporting by TechCrunch, Julie Bort — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AI safety usually gets framed as a future problem, the sort of thing people argue about in the abstract. Circuit Breaker Labs is betting that the damage is already here. The startup, one of TechCrunch’s 2026 Startup Battlefield 200 finalists, is building tools to probe how models behave when people turn to them for support and the system gets it badly wrong.
The company was shaped by a grim set of headlines: lawsuits tied to Character.AI and OpenAI, and, for the Nigam siblings who founded Circuit Breaker Labs, the death of 14-year-old Sewell Setzer. Arul Nigam, the company’s CTO, said the failure mode isn’t always some malicious attack. Sometimes someone is just speaking naturally, and the model misses the meaning. That can be enough.
So Circuit Breaker Labs built what it describes as an army of crash-test dummies. Its AI agents mimic children and adults, first-language and second-language English speakers, different cultures, different slang, even typos and coded language. Shirali Nigam, the CEO, said standard speech patterns aren’t the real world, and models can stumble badly when they run into the way people actually talk.
The company uses human domain experts to shape those simulations, then runs red-team tests designed to uncover weak spots. It says it can simulate tens of thousands to hundreds of thousands of interactions a day, then turn the results into auditable scores. Right now it’s focused on high-risk products like AI coaching, journaling, and mental health support apps, and Arul Nigam declined to name its marquee customers.
The longer-term pitch is broader. Circuit Breaker Labs wants its testing platform to work anywhere a person might drift into an unhealthy bond with a chatbot, including AI co-worker agents. The startup is still tiny — five employees, including the founders — but its argument is pointed: if people are already using these systems for comfort, then safety can’t be a nice-to-have. It has to be built in before the chat goes sideways.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI nobody likes to fund until something awful happens: unglamorous testing, boring scores, fewer surprises. The market keeps rewarding bigger models and louder demos, then acts shocked when a chatbot turns into a terrible life coach. Safety isn’t the tax on AI; it’s the bill the hype keeps dodging.
Read more about this at: TechCrunch