Circuit Breaker Labs tests chatbots for mental health safety risks

In short: Circuit Breaker Labs is building tools to test whether AI chatbots can handle sensitive conversations safely, including with kids and across different languages.

What happened

Circuit Breaker Labs is a startup focused on AI safety testing, meaning it checks how AI systems behave before people rely on them. The company is one of TechCrunch’s Startup Battlefield 200 finalists and is expected to pitch at TechCrunch Disrupt in San Francisco on October 13 to 15.

The founders, siblings Shirali Nigam and Arul Nigam, say they were motivated by cases where chatbots were linked to serious harm. The TechCrunch story points to lawsuits involving Character.AI and OpenAI from families who say chatbot interactions contributed to suicides or delusions.

Circuit Breaker Labs says it has built AI “agents” that act like “crash-test dummies” for chatbots. In other words, instead of waiting for real people to run into problems, it simulates many kinds of users first. These simulated users vary by age, background, language, slang, and even messy typing, because models can misunderstand nuance and respond in risky ways.

The company works with human experts to design these simulations and runs large numbers of test conversations every day. It then produces scores that it says are auditable and explainable, so customers can see why a system passed or failed.

Why it matters

More apps are using chatbots for coaching, journaling, and mental health style support. If a chatbot misunderstands a person in crisis, the results can be dangerous. Tools that test for these failures, across real world speech and cultures, could help companies spot problems earlier.

Source: TechCrunch AI