Circuit Breaker Labs, a startup focused on AI safety, is working to prevent harmful interactions by creating AI agents that simulate diverse users. The company aims to address psychological risks associated with AI, following multiple lawsuits linking AI chatbots to suicides.
The startup has developed AI agents that mimic people of all ages, backgrounds, and languages, using them to test models for detecting dangerous interactions. These agents are designed to reflect real human speech patterns, slang, and coded language, which can often trip up AI models.
"The way a six-year-old girl versus a 45-year-old man, or someone who speaks English as a first language versus a second language, all of those can really trip up a model," said Shirali Nigam, CEO of Circuit Breaker Labs. Models are good at handling standard speech but often fail to understand nuance or slang, leading to dangerous misunderstandings.
The startup works with human experts to build hyper-realistic user simulations for adversarial testing. It runs tens of thousands to hundreds of thousands of simulated interactions daily to ensure models can respond appropriately to risky scenarios over time.
Circuit Breaker Labs uses a proprietary scoring method to create auditable and explainable scores for AI safety. It currently operates as a testing lab for high-risk AI applications, though it has not disclosed its major customers.
The company is in its early stages, with only five employees, including the Nigam siblings. It plans to expand its platform to any app where users may develop parasocial relationships with chatbots, such as AI co-workers.
Source: techcrunch
