Asim, part of Y Combinator's Fall 2026 batch, is building standardized benchmarks for robotics teams that need to see how a robot behaves once it leaves a clean demo. A robot can finish a task in the lab and still fail when the object is unfamiliar, the load is heavier, the surface is slicker, a person enters the workspace, or an interaction goes wrong. The company runs those cases as repeatable tests on real robots, in the real world and in simulation, then gives engineers feedback on where performance breaks and whether the next software update actually fixes it. The scenarios test the capabilities underneath a completed task, including grasping, coordination, spatial reasoning, how parts fit together, dexterous manipulation, and adaptation when conditions change. Safety checks cover physical contact, including collision force when people enter the workspace, semantic mistakes where a harmless request produces an unsafe action, and whether the robot refuses instructions meant to cause harm. Teams receive detailed feedback on public audit scenarios, safety scores on private scenarios the model has not seen, and a comparison with a human baseline doing the same task in the same setting. Asim is also building a runtime safety layer that monitors a policy's actions during execution to anticipate risk before it becomes harm. The San Francisco company was founded by Douglas Tilley, Raphaël Lafargue, Tristan Lecourtois, and Maëlic Neau. Tilley, the CEO, is a PhD candidate in human-robot interaction and AI at the University of Bath. Lafargue, the CTO, holds a PhD in AI and was a senior research scientist at Sony. Lecourtois, the COO, previously worked at NASA JPL and InstaDeep. Neau is a two-time RoboCup@Home world champion with a PhD in computer science from Flinders University.