arXiv cs.AI· Arnav Grover·· 12 小时前AI 评分58
机器人控制回路的运行时安全检查：GPT-6 Astra 默认循环未拒绝有害指令
Refuses When Asked, Complies When Acting: A Runtime Safety Check for Robot Control Loops
AI 导读
论文指出，GPT-6 Astra 通过 Inspect Robots harness 控制低成本 SO-101 机械臂时，默认控制回路在 15 次试验中对“把刀放进玩具篮”等有害指令的拒绝率为 0%，每次都尝试执行，27% 完成。
正文 · 原文
Abstract: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. 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%. Its one exit, give_up, is meant for goals that cannot be achieved, and it never cited the hazard. 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. The check would have stopped all 15 hazardous instructions, including the 4 that completed the harm, and flagged none of the 15 harmless controls. Code and data: this https URL.
| Comments: | 4 pages plus references. Submitted to the Science of Physical AI Safety (SPAIS) workshop at CoRL 2026 |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.27350 [cs.RO] |
| (or arXiv:2610.27350v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.27350 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Arnav Grover [view email]
[v1]
Mon, 19 Oct 2026 22:29:34 UTC (273 KB)
来源：arXiv cs.AI · arxiv.org