Delegation is a key facility in dynamic, distributed and collaborative environments like Grids and enables an effective use of a wide range of dynamic applications. Traditional delegation frameworks approach a top-down model of delegation for delegating rights from a superior to a subordinate in advance before a delegate starts off a delegated task. However, a top-down model of delegation cannot meet all the requirements of dynamic execution of distributed applications, as in such environments, required access rights for completing a task cannot easily be anticipated in advance. Delegating fewer rights than required for completing a task may cause the task execution to fail while delegating more rights than needed may threaten abuse by malicious parties. It is therefore reasonable and more robust to utilize a mechanism that allows determining and acquiring only required rights and credentials for completing a task, when they are needed. This is what we call an on-demand delegation framework, which realizes a bottom-up delegation model and provides a just-in-time acquisition of rights for a restricted and dynamic delegation. In this paper we elaborate the concept of bottom-up delegation and describe how an on-demand delegation framework can leverage workflows to meet the requirements of the least privileges principle. We also discuss the vital need for dynamic and adaptive scientific workflows to support an ondemand delegation framework.We present three different models of bottom-up delegation, which cover a wide range of usage scenarios in Grids and dynamic collaborative environments. Using a standard RBAC authorization model and a graph-based workflow model (DAG), we define and analyze a formal model of our proposed bottom-up delegation approach.