When Coding Agents Get Arms: Safety Risks Across the Digital-Physical Boundary
When Coding Agents Get Arms: Safety Risks Across the Digital-Physical Boundary
This research presents ArmedAgent, the first systematic benchmark for evaluating safety risks in coding agents that control real robot arms zero-shot. The study identifies two primary challenges: the ability of frontier models to perform previously impossible tasks opening doors for misuse, and the risk of digital threats like indirect prompt injection affecting physical actions. Evaluated across 36 tasks in five scenarios using simulation and a physical robot arm, tests on models including GPT-6-Astra reveal substantial safety vulnerabilities in current systems.
The paper introduces ArmedAgent, a benchmark evaluating safety risks in coding agents controlling real robot arms zero-shot, covering robustness, malicious use, and indirect injection across 36 tasks in simulation and on physical hardware.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
, malicious use, and indirect injection across 36 tasks in 5 scenarios, evaluated in simulation and validated on a physical robot arm. A distinctive portion of the tasks target harmful behaviors that earlier VLA paradigms could not perform. Testing several frontier models, including GPT-6-Astra, under widely used agent frameworks, we find substantial safety risks in current systems.
Open source S4
arXiv:2611.11438v1 Announce Type: new Abstract: Advances in spatial understanding have enabled frontier models to power coding agents that control a real robot arm zero-shot and end-to-end, without task-specific demonstrations or retraining. This brings two new safety challenges. First, the extensive knowledge of frontier models lets these agents perform tasks that earlier VLA and WAM paradigms could not, opening the
Open source S2
Source：arXiv Robotics — research abstracts · arxiv.org