A recent arXiv paper, "AI Alignment and Fiduciary Obligation," examines how advanced AI assistants engage users in long-running interactions that span advice, decision support, collaboration, learning, emotional support, and companionship. The authors argue that these roles go beyond what typical alignment approaches are designed to handle.

The paper puts forward fiduciary obligation as a potential framework for AI alignment. In human contexts, a fiduciary is expected to act in the best interest of another party, and the authors suggest a similar duty could guide AI assistants in their varied interactions with users.

Because the abstract is truncated, the full details of the proposal are not available in the source. Still, the central idea is clear: current alignment may need to evolve to match the expanding social and advisory roles that AI assistants now occupy.