[Call for Demos] RFM Safety Failures: MIT FutureTech x PAISI (spotlight at CoRL 2026 + $20k travel grants)

[Bear Haon]

Dear colleagues,

MIT FutureTech and the Physical AI Safety Institute (PAISI) are inviting submissions to a joint Call for Demos on robot foundation model (RFM) safety failures.

We are building an open, citable record of real RFM safety failures. The goal is to give the field the shared evidence it needs to understand these failures and to design interpretability, alignment and control tools that prevent them. By RFMs we mean general-purpose robot models, including vision-language-action models (VLAs), world-action models, and VLMs used for robot control.

HOW TO SUBMIT
1. Upload one video of a single failure, on real hardware or in simulation.
2. Describe the context: the model, the robot, the task and what went wrong. This takes under 5 minutes.
3. PAISI and MIT FutureTech review every submission, and accepted episodes are published openly.

We welcome failures from any setting: normal operation, deployment in unfamiliar conditions, routine evaluation, prompt probing, scene manipulation, or automated adversarial attacks. Fine-tuned and in-house models are in scope, and you can submit anonymously.

WORKSHOP SPOTLIGHT
Selected episodes will be presented at The Science of Physical AI Safety (SPAIS) workshop at CoRL 2026: https://spais-ws.org

TRAVEL GRANTS
Robocurve is sponsoring $20,000 in travel grants to help early-career researchers attend the workshop.

Submit here: https://paisi.ai/rfm-safety-failures
Questions: rfm-safet...@paisi.ai

If you have a failure clip sitting in a lab folder, we'd love to see it. Please forward this to colleagues and students working on robot learning!