ai-ethics-policy

name	ai-ethics-policy
description	Creates AI usage policies for businesses with transparency commitments, bias mitigation, and disclosure guidelines. Use when establishing responsible AI practices.
allowed-tools	Read Write Glob
metadata	{"author":"matthewhitcham","version":"1.0"}

AI Ethics Policy

When to Use This Skill

Use this skill when you need to:

• Create an AI usage policy for your business or team
• Define transparency and disclosure guidelines for AI-generated content
• Establish quality standards and review processes for AI outputs
• Build trust with customers through responsible AI practices

DO NOT use this skill for technical AI safety research, legal compliance audits, or data privacy policies (consult a lawyer). This is for practical business AI ethics policies.

---

Core Principle

AN AI ETHICS POLICY IS A TRUST DOCUMENT — IT TELLS YOUR CUSTOMERS, TEAM, AND STAKEHOLDERS EXACTLY HOW YOU USE AI, WHAT YOU REVIEW, AND WHERE HUMANS REMAIN IN CONTROL.

---

Phase 1: Brief

Required Inputs

Input	What to Ask	Default
Business type	"What does your business do?"	No default — must be provided
AI usage areas	"Where do you use AI? Content, customer service, analysis, operations?"	No default — list all areas
Customer sensitivity	"How sensitive is your audience to AI usage? Very, moderate, low?"	Moderate
Disclosure stance	"Do you want to disclose AI usage publicly, internally only, or both?"	Both
Industry regulations	"Are there industry-specific AI guidelines you need to follow?"	None specific

GATE: Confirm the brief before drafting the policy.

---

Phase 2: Structure

Policy Sections

1. Purpose — Why this policy exists
2. Scope — What it covers (tools, use cases, people)
3. Approved Uses — Where AI is used and how
4. Prohibited Uses — Where AI must not be used
5. Human Oversight — Review and approval requirements
6. Disclosure Standards — When and how to disclose AI usage
7. Quality Standards — Accuracy, bias, and output requirements
8. Data Handling — What data is shared with AI tools
9. Review and Updates — How the policy evolves

GATE: Confirm the structure before writing.

---

Phase 3: Write

Purpose Statement

"[Business Name] uses AI tools to enhance productivity, improve quality, and serve our customers better. This policy ensures we use AI responsibly, transparently, and with human oversight. AI assists our work — it does not replace our judgment, expertise, or accountability."

Approved Uses

Document each approved use case:

## Approved AI Uses

| Use Case | AI Tool | Human Review Required | Disclosure Required |
|----------|---------|----------------------|---------------------|
| Content drafting | Claude/ChatGPT | Yes — before publishing | Yes — per disclosure standard |
| Email drafting | Claude/ChatGPT | Yes — before sending | No (internal communication) |
| Data analysis | AI analytics tools | Yes — verify conclusions | Depends on context |
| Image generation | DALL-E, Midjourney | Yes — before publishing | Yes — label as AI-generated |
| Customer support | AI chatbot | Escalation path to human | Yes — disclose bot usage |
| Research and summarization | Claude/ChatGPT | Yes — verify facts | No |

Prohibited Uses

## AI Must NOT Be Used For

- Making final decisions on hiring, firing, or compensation
- Generating legal advice or contracts without attorney review
- Creating content that impersonates a specific real person
- Processing customer data through unapproved AI tools
- Replacing human judgment on safety-critical decisions
- Generating content on sensitive topics without expert review
- Fabricating testimonials, reviews, or social proof

Disclosure Standards

## When to Disclose AI Usage

**Always disclose:**
- Blog posts, articles, or marketing content substantially written by AI
- AI-generated images used in marketing or products
- Customer-facing chatbots or automated responses
- AI-assisted analysis included in client deliverables

**Disclosure format:**
- Website footer or about page: "We use AI tools to assist with content creation. All AI-generated content is reviewed and approved by our team."
- Per-piece disclosure: "This [article/image/analysis] was created with AI assistance and reviewed by [Name]."
- Chatbot: "You are chatting with our AI assistant. A human team member is available if you need them."

**No disclosure required:**
- Internal notes and drafts
- Research and brainstorming
- Spell-checking and grammar tools
- Calendar scheduling and administrative automation

Quality Standards

## AI Output Quality Requirements

1. **Accuracy:** All facts, statistics, and claims must be verified by a human before publishing
2. **Originality:** AI output must be reviewed for plagiarism and substantially edited for voice
3. **Bias check:** Review outputs for gender, racial, cultural, or other biases before use
4. **Brand voice:** AI drafts must be edited to match brand tone and style
5. **No hallucinations:** Any specific claims, quotes, or data points must be fact-checked against primary sources

---

Phase 4: Polish

1. Data Handling Guidelines

## Data and AI Tools

**Do not input into AI tools:**
- Customer personal data (names, emails, financial info)
- Proprietary business data without approval
- Passwords, API keys, or security credentials
- Confidential client information

**Approved for AI input:**
- Publicly available information
- Your own original content and ideas
- Anonymized or aggregated data
- Internal process documentation

2. Policy Review Process

• Review the policy quarterly
• Update when new AI tools are adopted
• Solicit feedback from team members on practical challenges
• Monitor regulatory developments in your industry
• Document any incidents and update prohibited uses accordingly

3. Quality Checklist

## AI Ethics Policy Checklist

- [ ] Purpose statement explains why the policy exists
- [ ] All current AI use cases are documented with tools and review requirements
- [ ] Prohibited uses are explicitly listed
- [ ] Disclosure standards specify when, where, and how to disclose
- [ ] Quality standards require human review and fact-checking
- [ ] Data handling guidelines define what can and cannot be shared with AI tools
- [ ] Policy is accessible to all team members
- [ ] Review cadence is set (quarterly minimum)
- [ ] Customer-facing disclosure language is ready to publish
- [ ] Policy is practical and followable (not just aspirational)

---

Example

Business: Marketing consultancy

Customer-facing disclosure: "At [Agency], we use AI tools as part of our creative process. AI assists with research, drafting, and analysis — but every deliverable is strategized, reviewed, and refined by our human team. We believe in transparency: if you ever want to know how AI was involved in your project, just ask."

Internal policy excerpt: "All client deliverables that include AI-generated content must be reviewed by a senior team member before delivery. The reviewer confirms: (1) facts are verified, (2) content matches the client's brand voice, (3) no AI hallucinations or fabricated data, and (4) disclosure requirements are met per the client agreement."

---

Anti-Patterns

• Policy nobody reads — if the policy is 20 pages of legalese, nobody will follow it. Keep it practical and under 3 pages.
• Blanket "we use AI" without specifics — vague disclosure builds suspicion. Be specific about where and how AI is used.
• No prohibited uses — without clear lines, someone will use AI inappropriately. Define the boundaries.
• Ignoring data privacy — pasting customer data into AI tools without consent is a liability. Set clear data handling rules.
• Set and forget — AI capabilities change rapidly. A policy written 6 months ago may not cover new tools or use cases.

---

Recovery

• Team member violates the policy: Treat it as a learning opportunity. Review the incident, update the policy if needed, and retrain.
• Customer asks if content is AI-generated: Answer honestly. Transparency builds more trust than deflection.
• Industry issues new AI guidelines: Update the policy immediately. Note the regulatory source and adjust approved/prohibited uses.
• Policy feels too restrictive: Review whether restrictions are based on real risks or theoretical concerns. Loosen where the risk is low.