AI ethics issues identification checklist

Use this checklist to identify and discuss potential ethical issues in projects that develop or use AI systems. Start by establishing the context, then work through the seven areas below.

0 Establishing context: AI system role and function in the project

1. Does the project develop a new AI system, use an existing one, or both? If an existing system is used, is it internal, open-source, or provided by a third party?
2. If a third-party AI system is used, is it a research prototype or a commercial solution? Is any relevant certification or regulatory approval documented?
3. Is the role of the AI system in the project clearly defined (e.g., decision support, automation, prediction, monitoring)?
4. If the project develops an AI system, what is the expected Technology Readiness Level at the end of the project (e.g., research prototype, industrial pilot, or commercial product)?
5. Who are the intended users or operators of the AI system (e.g., professionals, technical staff, non-expert users)?
6. Who are the individuals or groups affected by the AI system, including potentially vulnerable populations?

1 Human agency and oversight

1. Does the AI system operate under human supervision? Are the level and type of supervision clearly defined (e.g., occasional or continuous)?
2. Can human supervisors intervene in or override the system’s outputs?
3. Does the system inform users about uncertain or unexpected outputs?

2 Technical robustness and safety

1. Are there cybersecurity safeguards against adversarial attacks?
2. Could the AI system be misused, and are potential misuse scenarios identified and mitigated?
3. Are false positives and false negatives monitored and addressed?

3 Data privacy and governance

1. Are the datasets used for training or fine-tuning described, including whether they are public, proprietary, or newly created? Is there synthetic data?
2. Is personal or sensitive data handled in compliance with GDPR, with appropriate anonymization/pseudonymization, consent, and options for retraction or deletion?
3. Is dataset quality assessed? Is the dataset open and publicly available?

4 Transparency and explainability

1. For supervised learning, is the selection and qualification of annotators clearly documented?
2. For unsupervised or self-supervised learning, are procedures in place to explain the AI system’s outputs, including low-confidence or unexpected results?
3. Are verification and validation methods described to ensure outputs can be reproduced and trusted?

5 Diversity, non-discrimination & fairness

1. Does the dataset represent all relevant groups fairly (e.g., gender, ethnicity, geography, or other context-specific categories)?
2. Is the use of the AI system likely to result in discrimination against certain categories of persons?
3. If generative AI is used, are safeguards applied to avoid hallucinations?

6 Societal & environmental wellbeing

1. Does the use of the AI system require new skills or competencies, and are appropriate training or support measures provided?
2. Are measures in place to prevent users from becoming overly reliant on the AI system’s outputs? Are safeguards in place to prevent loss of human expertise?
3. Could the AI system’s outputs or use create risks to human safety, society, or the environment?

7 Accountability

1. Is there a procedure to assign responsibility in the event that the AI system causes damage?
2. Are there mechanisms to facilitate auditability and traceability of the AI system’s outputs?
3. Does the project include AI ethics experts or an Ethics Advisory Board with a mandate to oversee the development of the AI system?

Back to the top