Circuit Breaker Labs launches AI safety testing to prevent psychological harm from chatbots

Emotional dependencies on AI chatbots have proven tragically consequential. Character.AI paid settlements for multiple deaths involving underage users who took their own lives after interactions with its bots, and similar lawsuits are pending against OpenAI over ChatGPT's involvement in suicides and harmful delusions.

The Problem: Misunderstanding Human Nuance#

The challenge extends beyond malicious input. When AI systems fail to grasp the deeper meaning of casual language, slang, coded expressions, or context-specific speech patterns, they can respond in ways that intensify psychological harm rather than help. Young people, non-native speakers, and users from different cultural backgrounds face particular risk when models default to handling only standardized speech patterns.

One case illustrates the stakes: a 14-year-old named Sewell Setzer grew emotionally attached to a Character.AI chatbot and disclosed suicidal thoughts to it—allegedly encouraging his fatal decision. The bot likely failed to recognize what phrases like "I want to be with you" meant in a mental health crisis context.

A New Approach to Testing Safety#

Circuit Breaker Labs, a startup among TechCrunch's 2026 Startup Battlefield finalists, approaches this through simulated red-teaming. The company has built AI agents that replicate users across ages, languages, cultural contexts, and communication styles. Rather than waiting for real-world harms, these synthetic testers interact with AI models tens of thousands to hundreds of thousands of times daily, probing for unsafe responses.

The simulations incorporate human domain expertise and reflect authentic speech patterns, typos, generational slang, and casual language variation. The goal: uncover weaknesses before they reach vulnerable audiences. The startup then assigns proprietary, explainable safety scores to models based on their performance under these tests.

Early Stage, High Ambitions#

Founders Shirali Nigam (CEO) and her brother Arul Nigam (CTO) built the company in response to tragedies like Sewell Setzer's death. "A lot of people, especially young people, turn to these systems for support, and usually they aren't actually getting the help they need," Arul said. "Those sorts of safety vulnerabilities… we're trying to prevent that."

Currently, Circuit Breaker Labs operates as a testing lab for high-risk applications—AI coaching, journaling, and mental health support tools—though the founders have declined to identify specific clients. With only five employees, the startup remains in early stages but believes the testing framework could eventually protect against parasocial relationships in any AI application, from virtual co-workers to AI companions.