Is your AI stack building Babel or Jerusalem?

# Human-First AI Policy for Team

**For:** Team (2-10 people)
**Tools in scope:** ChatGPT, Claude, Gemini, Copilot, image and transcription tools
**Version:** Standard starter policy

> This is an ethical governance starter for team discussion. It is not legal advice, compliance certification, ISO certification, EU AI Act compliance, NIST/OECD/UNESCO endorsement, or Vatican/Holy See endorsement. Adapt it with qualified advice for your context.

## Risk Classification for Selected Uses

| Use case | Risk | Required guardrail |
|---|---:|---|
| Writing/editing | Low | Allowed with ordinary human review and truth-checking. |
| Research | Medium | Verify claims against primary or authoritative sources. |
| Meeting summaries | Low | Allowed with ordinary human review and truth-checking. |
| Strategy | Medium | Human review required before external or consequential use. |
| Policy documents | Medium | Human review required before external or consequential use. |

## 1. Purpose

This policy lets this team use AI without surrendering human judgment. AI may assist the work, but it does not become the author, authority, caregiver, teacher, manager, pastor, lawyer, doctor, accountant or decision-maker.

The operating principle is human-first: AI as assistant, not authority; human care, skill, authorship, responsibility and accountability remain primary.

## 2. Allowed Uses

AI may assist brainstorming, drafting, editing, summarization, translation, coding assistance and research organization.

Selected lower-risk uses are writing/editing, research, meeting summaries, strategy and policy documents when a human reviews the result before it is shared or relied on.

Staff may use AI to improve clarity, generate options, organize notes, compare alternatives and prepare first-pass materials, provided the final output is reviewed by a human before use.

## 3. Restricted Uses

Restricted selected uses: Research (Medium); Strategy (Medium); Policy documents (Medium). These require human review before use outside the drafting process.

High-stakes uses require human review, documentation of the AI role, sources or inputs used, the human reviewer and an appeal or reconsideration path for affected people.

Selected decision impacts: reputation. Any AI use affecting these areas is treated as at least Medium risk and usually High risk.

## 4. Prohibited Uses

Fully automated consequential decisions about people are prohibited. AI must not make final decisions about grading, hiring, firing, discipline, benefits, eligibility, health, safety, legal rights, pastoral care or access to essential services.

AI must not be used for deception, impersonation, hidden persuasion, illegal discrimination, harassment, sensitive-data misuse, or surveillance/biometrics/emotion inference for evaluation.

Any use where no competent human can review the output is prohibited until review capacity exists.

## 5. Human Accountability

A named human owns every AI-assisted output. The human owner is responsible for accuracy, fairness, tone, confidentiality, safety and consequences.

AI output may not be cited as the reason for a decision. The responsible person must be able to explain the decision in ordinary human terms and reconsider it when challenged.

## 6. Truth and Verification

AI is not a source. Factual claims, quotes, citations, statistics, legal claims, code, formulas and calculations must be verified before use.

For research or policy work, use primary or authoritative sources where possible. For code and formulas, run tests or independent checks before relying on the result.

## 7. Disclosure

Disclose AI assistance on public-facing, stakeholder-facing, high-stakes or trust-dependent work.

AI chatbots and automated responders must identify themselves and offer escalation to a human. Synthetic or materially AI-altered media must be labeled when trust, consent, safety, reputation or public understanding matters.

## 8. Privacy and Data

Data in scope: internal. Stakeholders in scope: staff and clients.

Do not enter confidential or personal data into unapproved public AI tools. Use minimum necessary data and remove identifying details when practical.

If jurisdiction is unknown, treat privacy, employment, education, health, finance and legal-rights use as higher risk until reviewed.

## 9. Learning and Skill Preservation

AI should preserve human skill, authorship, judgment, care and accountability. For important work, the human should form an initial judgment or outline before asking AI to improve, challenge or organize it.

The team should keep human practice in the loop: reading source material, writing original drafts, checking calculations, explaining decisions and learning from mistakes rather than outsourcing them.

## 10. Governance and Review Cadence

This policy should be reviewed every six months, and sooner after a major incident, new tool, new sensitive data use, new high-stakes workflow or material change in law or guidance.

The review owner is a peer reviewer. The review should update approved tools, prohibited uses, examples, disclosure language, training needs and the risk classification above.

No selected use case is currently classified as High risk, but new consequential uses must be classified before launch.