Agent scope

An agent scope is an operator’s declaration of what an AI agent is for: its purpose, the tools it may use, and the limits it runs under. Declared before the agent works, it gives every subsequent security decision context, and turns "outside the declaration" into a detectable event.

Scope does two jobs with one declaration. It supplies context that reduces false positives (a database agent doing database work stops looking like an attack), and it enforces a boundary (a scoped agent calling an undeclared tool is held for approval). We measured the first effect on our production system on the ADR-Bench benign corpus (2026-08-28): benign sessions interrupted fell from 17.6% anonymous to 1.1% scope-declared, benign messages flagged from 0.74% to 0.12%. The full measurement covers methodology. The operator declares; the governance layer enforces what was declared.

Related

• Agent scope docs
• The measurement (blog)
• All glossary terms