Ethics becomes concrete when it becomes a checklist. It does not replace professional judgement, but ensures critical questions are not skipped in the sprint rush. Here is a framework you can adapt for your team — call it "AI UX Ethics Review".
Transparency: does the user know there is AI? Do they know what it does and does not do? Are limits communicated? Is disclosure contextual (not just footer)? If something is generated, is it clearly indicated?
Consent and control: explicit opt-in or opt-out? Granular control? Ability to disable? Data used for AI declared in the flow (not just privacy policy)? Right to correction and error reporting?
Fairness and inclusion: tested with diverse users by age, disability, language, socioeconomic context? Equity metrics monitored? Fallback for failed recognition/classification? Non-blaming error messages?
Accessibility: AI output validated for WCAG? Alt text, structure, contrast, plain language verified? Test with screen reader and keyboard-only?
Safety and worst case: what happens if the model hallucinates, discriminates, is used out of context? Is there an off-switch? Human escalation? Incident response plan communicated to the design team?
Document every review — even briefly. When something goes wrong post-launch, having a record of "we asked X, the team decided Y" protects the designer and improves future processes.