Responsible AI Guidelines for Public Health

Practical guidance for using AI in ways that are narrow, reviewable, source-aware, and accountable to human judgment.

Public health teams do not need vague promises about AI transformation. They need practical workflows that make repetitive work faster without weakening trust, clarity, privacy, or accountability.

The operating philosophy

Backyard-AI is built around a simple idea: AI should support public health work, not replace professional judgment. The safest use cases are usually narrow, well-scoped, source-aware, and reviewed by humans before important outputs are used or shared.

These guidelines can help teams discuss where AI belongs, where it does not, and what review steps are needed before AI-assisted work is used.

Core principles

  • AI should support public health work, not replace expert judgment.

  • Narrow workflows are safer and more useful than broad, ambiguous ones.

  • Important outputs need visible human review.

  • Claims should stay within the limits of the source material.

  • Early pilots should avoid sensitive data whenever possible.

  • Staff should know what a workflow will do and what it will not do.

  • Teams should document review steps when outputs may affect public-facing communication, planning, reporting, or decision support.

What Backyard-AI is for

Backyard-AI is designed to support:

  • Evidence synthesis and summary support

  • Health communication drafting and review support

  • Workforce training and teaching support

  • Reporting and workflow support

  • Internal knowledge and policy support

  • Resource navigation and approved-source knowledge assistants

  • Responsible implementation planning

Default safeguards

Human review

Important outputs should be reviewed by a person who understands the topic, context, audience, and intended use.

No hallucinated citations

If a claim cannot be supported by the available source material, it should be flagged, revised, or removed.

No sensitive data in public demos

Early demos and examples should rely on public, synthetic, de-identified, or explicitly approved materials.

What should not be pasted into a public AI model

  • Protected health information

  • Personally identifiable information

  • Restricted personnel information

  • Legally protected case information

  • Confidential organizational records

  • Unapproved grant, contract, or internal strategy materials

  • Any data that your organization would not allow in an external web tool

Source boundaries

Factual workflows should use a narrow approved-source list whenever possible.

No overclaiming

AI-assisted outputs should not sound more certain than the evidence supports.

Traceability

When appropriate, teams should keep a record of the source material, prompts, instructions, reviewer, and approved output.

Better early-pilot inputs

  • Public guidance documents

  • Synthetic examples

  • De-identified text

  • Approved uploaded documents

  • Public reports

  • Training scenarios

  • Sample data created for demonstration

Risk tiers

Lower-risk workflows

Examples: meeting summaries, internal action lists, training handout drafts, brainstorming, plain-language rewrites, and non-sensitive email drafts.

Review expectation: staff review before use.

Review checklist

Moderate-risk workflows

Examples: dashboard narratives, guidance comparisons, public-facing communication drafts, resource guides, grant narratives, and program summaries.

Review expectation: subject-matter review, source check, and approval before sharing.

Before using an AI-assisted output, ask:

  • Is the workflow using approved source material?

  • Was any sensitive data included?

  • Does the output overstate certainty?

  • Are dates, names, numbers, citations, and contact details correct?

  • Is the output within the intended scope?

  • Is the language appropriate for the audience?

  • Could the output create confusion, stigma, bias, or harm?

  • Has the right person reviewed and approved it?

Disclosure examples

Internal disclosure option: This draft was prepared with AI assistance and reviewed by a human before use.

Public-facing disclosure option: This material was drafted with AI assistance using approved source material and reviewed by a human prior to release.

Workshop or training disclosure option: This activity includes AI-assisted examples for learning purposes. Outputs should be reviewed before use in real-world settings.

Need help applying responsible AI guidelines to your team?

Backyard-AI can help you turn these principles into practical guidance, review checklists, staff training, risk tiers, and workflow-specific expectations.

Higher-risk workflows

Examples: workflows affecting legal compliance, clinical interpretation, sensitive data handling, regulatory decisions, public alerts, or high-consequence public communication.

Review expectation: do not proceed without explicit organizational approval, appropriate expertise, and a documented risk review.

These guidelines are for general planning and training purposes. They are not legal, clinical, compliance, or procurement advice. Organizations should follow their own policies and consult appropriate experts when working with sensitive data, regulated settings, or high-consequence decisions.