Michael Schroeder

The distinction matters particularly when a field's new systems inherit problems that earlier work already helped to define.

Jerome H. Saltzer and Michael D. Schroeder's nineteen seventy-five article, The Protection of Information in Computer Systems, provides a concrete case.

Two, when an organization allows an AI agent to alter a record or invoke a service, it grants authority over resources for which someone remains accountable.

The model may be new, but the organization still needs to establish why an action is permitted, on whose behalf it is performed, and what prevents the agent from exceeding its task.

A single age limit cannot perform both judgments.

How current research applies the principles.

Kaiyun Zhang and colleagues explicitly return to Saltzer and Schroeder in their twenty twenty-five position paper preprint, LLM agents should employ security principles.

Their agent sandbox framework applies least privilege and complete mediation to agent data access and communication.