Back to the AI Control Library
11 controls | 11 core | 0 supplemental
Why this family matters
A broad instruction to be helpful does not define an acceptable operating boundary. This family addresses how an AI system's purpose, permitted activity and limits are translated into enforceable constraints. It connects organizational policy to input handling, output handling, resource use and the scope of the system's work.
What this family covers
The controls cover objectives, validation and filtering, system-prompt protection, prohibited uses, scope restrictions, rate and resource limits, output structure, confidence thresholds and policy-as-code. These controls provide boundaries that other safeguards can monitor and enforce.
How the four control types apply
Technical controls enforce scope, validation, filtering and resource limits. Operational controls establish how objectives and policies are defined, approved and maintained. Governance controls establish applicable external prohibitions and obligations. Contextual limits shape how confidence and output information guide human use.
Where to begin
Write down what the system may do, what it must never do, and which technical mechanisms enforce those limits.
Controls in this family
This family contains 11 controls: 11 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
- CF11-001: AI System Objective & Scope Definition
- CF11-002: AI Input Validation & Filtering
- CF11-003: AI Output Validation & Filtering
- CF11-004: System Prompt Isolation & Protection
- CF11-005: Prohibited Use Case Technical Enforcement
- CF11-006: EU AI Act Prohibited Practice Controls
- CF11-007: AI System Scope Restriction
- CF11-008: AI System Rate Limiting & Resource Constraints
- CF11-009: AI Output Format & Structure Enforcement
- CF11-010: AI Confidence Threshold Enforcement
- CF11-011: Policy-as-Code for AI Constraints
Applying this family
Define the objective and permitted scope before selecting technical controls. Enforce data and action boundaries independently of model instructions, validate outputs for their actual consumers and limit cumulative resource use. Treat uncertainty as a routing decision and test policy rules against bypass, delegation and unavailable enforcement.
Version 1.0