What's this about?
This paper maps the risks of handing real work to AI agents in organizations — focusing not on capability limits but on accountability gaps (who's responsible when an agent acts) and human skill atrophy (people losing abilities they've delegated away).
What exactly is "delegation risk"?
When you hand a task to an agent, you also hand off some control and understanding. Delegation risk is the downside of that: decisions get made with no clear human owner, and over time the humans get worse at the delegated skill, making oversight harder. It's an organizational and social risk, not a technical bug.
Why is this notable?
It reframes the agent-safety conversation around governance and human factors — exactly the "who's liable?" question today's gym hack raises. As companies deploy agents at work, these are the risks that show up in practice.