Subjects, roles, resources, purposes, and grants form an authorization graph. Retrieval should enforce allowed paths before content enters model context, not only before final display.

Authorization Can Be Evaluated as a Graph

Agent identity, messages, tools, and shared memory may be compromised or mistaken. Authenticate senders, authorize actions, validate payloads, and minimize transferred data at every boundary.

Every Agent Boundary Is a Trust Boundary

Map collection, inference, storage, retrieval, sharing, logging, and deletion across graph edges. Threats include overcollection, linkage, memorization, unauthorized inference, and retention drift.

Privacy Threat Modeling Follows Data Paths

Injection succeeds when untrusted content is allowed to alter privileged instructions or actions. Every ingestion, memory, message, and tool-return edge must preserve trust labels and authority.

Prompt Injection Is a Graph-Level Data-Control Confusion

In this situation—preventing a tutor from retrieving another student's accommodation record—which choice best applies “Authorization Can Be Evaluated as a Graph”?

Learn Before

Related

Related