AI & ML New Capability

ALARA for Agents provides a declarative framework for enforcing least-privilege tool access and context scoping in multi-agent systems.

March 24, 2026

Original Paper

ALARA for Agents: Least-Privilege Context Engineering Through Portable Composable Multi-Agent Teams

Christopher J. Agostino, Nayan D'Souza

arXiv · 2603.20380

The Takeaway

It introduces a structured 'Context-Agent-Tool' (CAT) layer that guarantees behavioral changes via file structural parsing rather than prompt suggestions. This is a critical step for deploying multi-agent systems in security-sensitive enterprise environments.

From the abstract

Industry practitioners and academic researchers regularly use multi-agent systems to accelerate their work, yet the frameworks through which these systems operate do not provide a simple, unified mechanism for scalably managing the critical aspects of the agent harness, impacting both the quality of individual human-agent interactions and the capacity for practitioners to coordinate toward common goals through shared agent infrastructure. Agent frameworks have enabled increasingly sophisticated