Technology

How Do You Make an AI Chatbot Safe for Your Kids?

Martin HollowayPublished 2month ago2 min readBased on 2 sources
Reading level
How Do You Make an AI Chatbot Safe for Your Kids?
Photo by X on Unsplash

Circuit Breaker Labs has been named one of TechCrunch's 2026 Startup Battlefield 200 finalists for work on making AI safer across languages and cultures. TechCrunch

The company was started by siblings Shirali Nigam, its chief executive, and Arul Nigam, its chief technology officer.

Its core idea is AI agents the company compares to an army of crash-test dummies. They pretend to be people of all ages, backgrounds, languages and cultures to test if models can spot dangerous, psychologically harmful chats.

The agents are built with human domain experts into very lifelike user simulations. The simulations are used for red-team tests, planned attempts to trick the model and find failures.

The company reports running tens of thousands to hundreds of thousands of simulated chats per day.

It works as a safety testing lab for high-risk uses such as AI coaching, journaling, or other mental health support apps. It grades model behavior with its own scoring method to make safety scores that can be checked and explained.

Its stated position is that “it independently stress-tests AI to find dangerous failures so real people never do.”

The broader context here is a change from testing for right answers to testing for safe behavior. Accuracy tests check if a model answers correctly. Safety testing for coaching and support checks if it stays safe over long, emotional chats in many languages where harm grows slowly.

In my view, the hard part is culture coverage and clear fixes. Users of different ages, languages and cultures test safety rules better than tests from one place. Results depend on lifelike simulations, on copying rewording and quiet signs of distress, and on linking each failure to one chat engineers can replay and fix.

Looking at what this means for teams building these companions, outside testing at scale changes the routine. Inside testers know the hidden instructions and refusal rules. Outside users do not. That gap helps. Scale also allows retesting after model or rule changes, not just once before launch. Clear scores matter because product, safety and legal teams need one shared record, not just pass or fail.

Worth flagging in that light is what this could allow at large scale. Crash testing did not end vehicle risk, but it created a repeatable way to find failure before sale. Used regularly for AI coaching and journaling, testing in many languages could let builders break, fix and retest systems before real users face the same failures.