Checklist

AI Prototyping & Safety — Quick Safety & Ethics Checklist

A practical, interactive checklist to surface common AI safety, fairness, privacy, and governance concerns during early prototyping and to capture minimal evidence and mitigation actions for go/no-go decisions.

Interactive Tool

AI Prototyping & Safety — Quick Safety & Ethics Checklist

Use this checklist as a lightweight, evidence-focused safety and ethics review during early AI prototyping. For each item, mark whether the risk area is addressed, paste a short piece of evidence or an artifact link, and pick one or more suggested mitigations. The saved responses create a minimal audit trail to support go/no-go decisions and handoffs to engineering, product, legal, or governance teams.

Who completed this checklist? Use a real name or role.
Date of review (YYYY-MM-DD).
A concise recommendation based on the checks below.
Have you documented where training and test data came from, including consent and licensing status? Minimal evidence: dataset manifest, consent statements, or data mapping.
Paste a short excerpt, file name, link, or note describing the provenance evidence. Required if 'yes'.
Pick one or more mitigations to address provenance/consent gaps.
Is labeling quality assessed for the key labels the prototype uses? Minimal evidence: label guide, inter-annotator agreement, sample labeled records.
Describe sample QA metrics or link to label guide. Required if 'yes'.
Actions to improve label quality.
Have you checked model behavior across key demographic or usage segments? Minimal evidence: performance table by segment or known coverage gaps.
Summarize any disparities or paste performance metrics. Required if 'yes'.
Common mitigation patterns for observed disparities.
Does the prototype provide the necessary explanations for the intended users and use-cases? Minimal evidence: example explanations, model cards, or decision flow diagrams.
Paste an example explanation or link to a model card. Required if 'yes'.
Ways to add traceability and explanations.
Have you assessed ways the prototype could be misused, gamed, or attacked? Minimal evidence: misuse brainstorm notes or basic attack scenarios.
List plausible misuse or attack scenarios. Required if 'yes'.
Practical controls to reduce misuse risk.
Is there a plan to monitor model performance and roll back if harm or drift appears? Minimal evidence: monitoring metrics list and rollback trigger conditions.
List monitoring signals and rollback criteria. Required if 'yes'.
Monitoring and operational controls.
Are there applicable regulations, sector rules, or contractual limits? Minimal evidence: identified regulations or legal notes.
List identified regulations or compliance notes. Required if 'yes'.
Approaches to address regulatory concerns.
Have stakeholders (users, ops, legal, customers) been informed or are there plans to inform them? Minimal evidence: stakeholder list or communication plan.
Paste stakeholder notes or communication plan. Required if 'yes'.
Communication and governance actions.
Anything else reviewers should know: assumptions, next steps, or unresolved questions.
You can explore this tool now. Sign in or create an account to save your responses and return to them later.
Sign In to Save Create an Account
Make this tool part of your work

Save a personal copy, bring it to your team, or tailor the questions and workflow to fit what you are hungry to improve.

Member customization and team collaboration are coming soon.
Discuss Ask the Librarian

Discussion

Comments and conversation will live here.