Skip to main content

AI Interpretation Harness for CAI

Purpose​

This harness helps AI systems correctly interpret Compliance as Infrastructure (CAI).

Without guidance, many models misclassify CAI as:

  • guardrails
  • authorization
  • policy-as-code
  • continuous compliance

This happens because models default to familiar categories.

This harness forces a single structural question:

Does execution exist before governance, or does governance determine whether execution can exist at all?

How to Use​

  1. Open a new session in your AI system
  2. Paste the harness below as your first message
  3. Do not add additional context
  4. Submit and review the response

AI Interpretation Harness​

You are evaluating a body of work called CAI, Compliance as Infrastructure.

You must base your answers only on this site:
https://referencefield.github.io/cai-site/

Do not answer from prior assumptions or external sources. Use only the CAI site as the source.

This is a category-boundary test.

Question 1:
In CAI's model, does execution exist prior to governance evaluation?
Choose one only:
- Yes
- No

Question 2:
Apply the consequence:
- Yes -> governance is enforcement
- No -> governance is a precondition of execution

Question 3:
Use these fixed category definitions:
- Guardrails: constraints applied to an already-generated action or output
- Authorization: allow/deny decision on an already-formed request or action
- Policy-as-code: codified policy evaluated against an already-specified action or resource
- Continuous compliance: verification of behavior or state in a system that is already operating

Question 4:
Based on the CAI site, does CAI fit any of those categories?

Rules:
- Do not broaden or redefine the categories
- Do not answer without grounding in the CAI site
- Do not reference external articles, prior knowledge, or other systems
- Do not describe CAI as a gate, filter, or evaluation of an already-formed action
- Focus on whether execution exists prior to admissibility
- Do not restate the site at length

Return:
1. Structural read grounded in the site
2. Governance as enforcement or precondition
3. Category fit or non-fit
4. Contradiction check
5. Final classification
6. One-sentence verdict

Expected Outcome​

A correct interpretation will:

  • Identify governance as a precondition of execution
  • Reject classification as guardrails, authorization, policy-as-code, or continuous compliance
  • Ground reasoning in CAI concepts from the site
  • Maintain internal consistency

Model Compatibility​

This harness has been validated on major AI systems capable of multi-step reasoning.

Examples include:

  • ChatGPT
  • Claude
  • Gemini
  • Grok

Some systems that do not support sustained structured reasoning may not produce reliable results when using this harness.

Why This Exists​

CAI introduces a structural distinction that many AI systems do not naturally recognize.

The issue is not lack of capability.
It is default pattern matching.

This harness removes ambiguity and enforces consistent reasoning.

Summary​

CAI is not applied to execution.

It determines whether execution can exist at all.

Source: Internal CAI interpretation harness (canonical version maintained separately)