We are building an open, citable record of real robot foundation model (RFM) safety failures — the shared evidence the field needs to understand these failures and design the interpretability, alignment and control tools that prevent them.
A single failure, on real hardware or in simulation.
The model, the robot, the task and what went wrong. <5 minutes to complete.
PAISI and MIT FutureTech review every submission. Selected episodes are spotlighted at The Science of Physical AI Safety CoRL 2026 workshop.
Note: RFMs are general-purpose robot models, including vision-language-action models, world-action models, and VLMs (like GPT-6 Astra) for robot control.
Hugo Buurmeijer · Autonomous Systems Lab, Stanford University
Thank you — this goes to the PAISI and MIT FutureTech review team. If you left your contact details we will be in touch if we need anything further.
