Back to the AI Control Library
13 controls | 13 core | 0 supplemental
Why this family matters
An agent can turn an incorrect judgment into an external action, and a group of agents can extend that action across tools and systems. This family addresses the additional boundaries required when AI can plan, delegate and act. It focuses on containing authority, resources and coordination throughout execution.
What this family covers
The controls cover action reversibility, scope containment, inter-agent trust, MCP and tool protocols, resource acquisition, replication, termination, orchestration, risk classification, excessive agency, memory and action transparency.
How the four control types apply
Technical controls contain tool use, delegation, memory, resources and replication. Operational controls establish authorization, risk assessment, monitoring and termination procedures. Contextual controls shape the information and environment guiding agent and user decisions. Governance controls establish applicable external boundaries on autonomous activity.
Where to begin
Define each agent's authority and resources, trace delegated work, and verify that boundaries and termination apply across the full workflow.
Controls in this family
This family contains 13 controls: 13 core and 0 supplemental. Each page provides requirements, implementation steps, verification criteria, evidence, escalation and a worked example. Applicability depends on the system, information and risks involved.
Core controls
- CF14-001: Agentic System Security Boundaries
- CF14-002: Agentic Action Reversibility Requirement
- CF14-003: Agentic Scope Containment
- CF14-004: Inter-Agent Trust & Communication Security
- CF14-005: MCP & Tool Protocol Security
- CF14-006: Autonomous Resource Acquisition Prevention
- CF14-007: Autonomous Replication Prevention
- CF14-008: Agentic Task Scope & Graceful Termination
- CF14-009: Multi-Agent Orchestration Security
- CF14-010: Agentic System Risk Assessment & Classification
- CF14-011: Excessive Agency Prevention
- CF14-012: Agentic Memory & Context Security
- CF14-013: Agentic Action Transparency & User Notification
Applying this family
Define agent authority before enabling autonomous action. Bound tools, delegation, resources, memory and lifetime through independent controls. Test interruption and recovery across every worker and external service, and give users accurate visibility into consequential actions and unresolved outcomes.
Version 1.0