Permissions vs. Policies in LLM Systems: What Matters Where

LLM agents call tools, read data, and act on behalf of users. Securing them depends on keeping two concepts apart: permissions and policies. A third element, prompt instructions, is often mistaken for