Should intelligent autonomous agents always obey human commands or instructions? 
We argue that, in some contexts, they should not.
Most existing research on collaborative robots and agents assumes that a “good” agent is one that complies with the instructions it is given and works in a predictable manner under the consent of the human operator it serves (e.g., it should never deceive its operator). The goal of this workshop is to challenge this assumption and to rethink the desired abilities and responsibilities of collaborative agents. These include, for example, exhibiting behavior that attempts appropriate and harm-preventing non-compliance (e.g., safety constraints in autonomous vehicles or training LLMs to avoid potentially harmful or norm-violating output), among others.

Submission website:

https://easychair.org/conferences/?conf=radai26