When State Becomes an Attack Surface: State-Semantic Injection in LLM-Driven Embodied Agents
The paper frames an embodied agent’s perception of state as a new place to inject attacks.
The authors argue that LLM-driven robots now reason over scene states, object attributes, spatial relations, and execution feedback, not just user text. That makes the environment itself part of the prompt surface when the model turns perception into plans. The paper is listed as submitted to USENIX Security 2027 and categorized under Robotics and AI. ArXiv · AI/CL/LG's note
The authors argue that LLM-driven robots now reason over scene states, object attributes, spatial relations, and execution feedback, not just user text. That makes the environment itself part of the prompt surface when the model turns perception into plans. The paper is listed as submitted to USENIX Security 2027 and categorized under Robotics and AI. ArXiv · AI/CL/LG's note
score 4