Microsoft Scout

Autopilot — always-on agent, Frontier preview. A governance case for what happens when a tool acts without being asked.

Row Edition use

Use as a thought drill: if an agent auto-triaged fire claims, drafted alerts, or watched the emergency inbox, which decision would it never own? Map named identity, containment, policy checks, and audit receipts to the six layers.

Open the fire-themed prompt library →

Human check before it ships

An always-on agent needs an explicit scope, named owner, runtime boundary, stop mechanism, and auditable record before it gets authority to act.

Source context: Computerworld reporting on Scout.

Bring it to the room

Use this tool with non-sensitive, public, or fictional material. The artifact is a draft; the approval, boundary, and receipt remain human.

Open the six-row checklist →