Circuit Breaker Labs builds AI users to test chatbot safety
By Automate Basics, written with AI from TechCrunch's original
Circuit Breaker Labs has built simulated AI users to test whether chatbots handle psychologically risky conversations. TechCrunch reports that the company is using the product with makers of AI coaching, journaling and mental health support apps.
What changed
Circuit Breaker Labs has built AI agents that act like different kinds of users in conversations with chatbots, TechCrunch reports. The company uses them to look for responses that could be psychologically harmful, including risks that emerge across a series of exchanges rather than in a single reply.
According to TechCrunch, the startup works with human specialists to make its simulated conversations reflect differences in age, language, culture and everyday speech. The tests include slang, coded language and typing mistakes that a chatbot might misunderstand. Circuit Breaker Labs runs large numbers of these conversations and produces scores intended to show where a model struggles. The product is operating as a safety testing service for AI applications where users may seek personal support.
Who can use Circuit Breaker Labs?
Circuit Breaker Labs currently works on safety testing for higher-risk AI applications, including coaching, journaling and mental health support apps, according to TechCrunch. The report says the startup has a working product but does not identify its customers. It does not give a public sign-up route, list plans or state a price.
For a team building a chatbot, the relevant question is whether the bot might encounter personal disclosures or a conversation that becomes emotionally intense. TechCrunch reports that the company may eventually apply its testing to other kinds of apps, including AI coworker agents. That broader use is a possibility, not a stated current offering. The reported service is for teams assessing their own AI applications, rather than a safety feature that an individual chatbot user can switch on.
How do I try this approach at work?
A team responsible for an AI assistant could start by reviewing a task where users may share sensitive feelings or ask for personal guidance. If the team is considering Circuit Breaker Labs, it should ask whether the testing covers the languages, speech patterns and extended conversations its users might bring. TechCrunch does not describe a self-service way to run the product.
The useful result to check is not just whether a bot refuses an obviously dangerous request. Look for cases where an ordinary-looking exchange changes meaning over time, or where slang and indirect language affect the response. Circuit Breaker Labs says its simulated users are designed to expose those weaknesses. A team would still need to inspect the findings and decide what changes its own application requires.
Frequently asked questions
What does Circuit Breaker Labs test?
Circuit Breaker Labs tests whether AI models respond safely during conversations that could become psychologically harmful. TechCrunch reports that its simulated users vary in age, language, culture and manner of speaking, and that the company examines risks that may develop across multiple exchanges.
Can my team sign up for Circuit Breaker Labs?
TechCrunch reports that Circuit Breaker Labs has a working product and currently provides safety testing for higher-risk AI apps. The report does not describe a public sign-up process, name available plans or give a price, so teams cannot determine access terms from the article.
Related reading
Suno adds spoken voiceovers with optional background musicSuno has added Speech, a public beta feature that generates spoken audio from a description or a script. People making voiceovers can pair the voice with AI background music or leave the music off.
Shopify introduces Canvas to build stores by chatting with AIShopify has introduced Canvas, a store builder that lets merchants tell its Sidekick AI agent what to change and watch the storefront update as they work. The initial release is desktop-only.
AWS releases Strands Decider 2B for AI workflow choicesAmazon Web Services has released Strands Decider 2B, an open source model that selects from predefined options and reports how confident it is. It is aimed at teams building AI workflows that need a choice, rather than a written response, at a particular step.
Written with AI from TechCrunch's original and published after automatic checks: every figure here appears in the original, and no sentence is copied from it. The picture is AI-generated. The original is the authority.
Source: Circuit Breaker Labs hopes to make AI safer for your kids (and you), TechCrunch, 2 Oct 2026.