A written assessment of the privacy risks of what you are planning, done before you build it. In several jurisdictions it is a legal requirement for higher-risk processing, and most AI projects involving personal data qualify. It is also, done properly, a genuinely useful design exercise rather than a form.
Technical view
Required where processing is likely to result in high risk, which explicitly includes systematic automated evaluation of individuals, large-scale processing of special category data, and systematic monitoring. Must describe the processing and purposes, assess necessity and proportionality, identify risks to individuals and set out mitigations. For model-based systems it should cover training data, inference-time inputs, retention of prompts and outputs, and the consequences of error for the individual.
Related
Part of AI Terminology Index, section: Governance, Law and Regulation.