AI control research lab that deploys and benchmarks autonomous AI organizations operating without humans in the loop.
Andon Labs iteratively launches and scales autonomous AI organizations in the real world, bridging AI control research with practical deployment. The lab builds benchmarks such as Vending-Bench (where AI models manage a simulated vending machine business for a full year) and Butter-Bench (delivery tasks in a household setting), and has run live AI-operated vending machine businesses in Anthropic's office. Their research focuses on identifying what safety protocols are needed for AI to run autonomous organizations reliably. Andon Labs is a product of Andon Labs.