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

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.