AI agents that act on a user's behalf by executing tool calls create a tricky problem: how should permissions be managed? The authors of a new arXiv paper argue that it is unclear what role users could or should play in that process, despite a range of proposed solutions.

To help answer this, they introduce Janus, a 'playground' for user-involved agentic permission management. The platform appears designed to let researchers experiment with different ways of involving users in permission decisions, rather than settling on a single approach.

The paper's contribution is to frame the question and provide a common space for testing ideas. By comparing approaches in a shared environment, it may become easier to see which permission models actually work in practice.