Analysis
Least Agency in AI Agents: Runtime Enforcement Guide
Publication date unknown · Discovered trussed.ai
Page published
Publication date unknown · First observed: 5 Nov 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Least-agency controls can limit the damage from an injected objective or mistaken agent decision by restricting what each action can do across connected systems.
Trussed's least-agency guide explains how to constrain AI agents' tools, action authority, autonomy and access to untrusted content at runtime. Recommended controls include per-call authorization, MCP connection approval, sandboxing and egress restrictions, reduced permissions for delegated tasks, human approval for high-impact actions and durable audit records.