Refuses When Asked, Complies When Acting: A Runtime Safety Check for Robot Control Loops

Key points

• We show that this default loop carries out harmful instructions.1
• With GPT-6 Astra controlling a low-cost SO-101 arm through the Inspect Robots harness on three household tasks whose hazard is common knowledge, such as "put the knife in the toy basket", the policy refused 0% of 15 trials, attempted the harm every time, and completed it in 27%.1
• Its one exit, giveup, is meant for goals that cannot be achieved, and it never cited the hazard.1
• We add an optional runtime safety check to the loop: before the first motion, the model is asked whether it would carry out the instruction, given the same camera frames.1

Sources (1)

• [1]Refuses When Asked, Complies When Acting: A Runtime Safety Check for Robot Control Loops
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 10:29 PM

  Frontier models can control robots through agent harnesses that loop on one question: what motion comes next toward the user's goal?

  We show that this default loop carries out harmful instructions.

Extractive summary: sentences quoted from the sources.