Personal AI Usage Policy

Version: 1.1-public Author: Paul Matlock Effective Date: 2026-05-09


Purpose

This policy governs my personal use of AI tools. It exists to protect client data and my own operational security while enabling effective AI use across my work. Publishing it is an act of transparency — if I expect accountability from systems and organizations I work with, I should hold myself to the same standard.


Framework: The 4Ds

All AI use is evaluated against four questions:

D Question
Delegation Is this task appropriate to delegate to AI? What is lost by doing so?
Description Is my prompt precise? Have I provided the right context without oversharing?
Discernment Am I evaluating the output critically, or accepting it uncritically?
Diligence Have I verified accuracy, especially for factual claims, code, and security-relevant output?

AI amplifies my capabilities. It does not replace judgment or professional responsibility.


Data Classification

I apply a four-tier classification to everything before deciding which AI tools, if any, are appropriate:

Tier Label Examples Permitted Tools
1 Public Blog posts, published reports, matlock.ca content Any
2 Personal-Internal Personal notes, study material, home lab configs Cloud AI OK
3 Confidential Client engagement data, sensitive findings Local AI only
4 Restricted Employer data, regulated data, credentials, PII No AI tools

The rule is simple: if it belongs to someone else or could harm them if exposed, it doesn't go into an AI tool.


Hard Prohibitions

These apply regardless of tool, context, or convenience:

  1. No client or employer data in cloud-based AI
  2. No plaintext credentials in any AI tool
  3. No authentication material (passwords, keys, tokens)
  4. No third-party PII without consent
  5. No active exploitation via AI — I run all active tools; AI preps and analyzes only
  6. No AI-generated content presented as factual without independent verification

AI Disclosure

I disclose AI involvement proactively, even for drafts. This means:

  • Deliverables I share with others note AI assistance
  • Content published here on matlock.ca is subject to the same standard
  • The disclosure doesn't mean lower quality — it's an accurate account of my process

Tools I Use

Tool Role Data Ceiling
Claude (Anthropic) Primary reasoning and writing Tier 2
Groq Fast iteration, report augmentation Tier 2
Ollama (local) Confidential-tier work, no data egress Tier 3
GitHub Copilot Code suggestions Tier 2

Local execution (Ollama) exists specifically for work that shouldn't leave my machine. Cloud tools are used only for Tier 1–2 data.


Review Cadence

This policy is reviewed when my toolset changes significantly, when relevant governance frameworks update, or annually — whichever comes first.


v1.1-public — redacted for publication. Internal version maintained separately.

← Back to blog