Responsible AI use begins before a prompt is sent. Decide whether the task is suitable, what data may be shared, how output quality will be reviewed, and who has authority to approve the result.
Minimize and protect input data
Provide only what the task requires. Avoid personal, confidential, or client information unless an approved tool and lawful, policy-compliant process are in place. Check retention, access, and training terms for the service being considered.
Check for uneven impact
Review whether an output makes unsupported assumptions about people, omits relevant perspectives, or produces different treatment across groups. Do not rely on a generic disclaimer to correct a flawed recommendation. Escalate consequential decisions to qualified people and established procedures.
Assign responsibility
Define a reviewer, a source-check process, a stop condition, and a way to report a problem. Respect rights and consent for source content and generated media. Human review should be capable of changing or rejecting the output, not merely approving it automatically.