Privacy, retention, and responsible AI

Use only the data needed for the task, and understand how it flows through prompts, traces, evaluation datasets, indexes, and backups. Apply data classification, access controls, retention limits, and deletion procedures consistently. Avoid copying real personal data into development examples when synthetic data will work. Review the provider and service settings that govern data processing before using sensitive information. Responsible deployment also considers harmful or biased outputs, accessibility, user disclosure, human oversight, and a route to report problems. Define who reviews incidents and how a model or index can be rolled back. Monitor representative outcomes rather than assuming a one-time approval remains valid after data or model changes. Map sensitive data across the full flow: original sources, chunks, prompts, model responses, traces, evaluation exports, and backups. Collect only what is necessary, document the purpose, restrict who can inspect it, and set deletion and retention behavior for each store. Use synthetic data for development where possible. Provide a clear way for users to report harmful or incorrect output, and define who triages incidents and can disable a feature. Revisit the assessment when the model, data source, user population, or intended use changes. Implementation drill: trace one representative request through the stages in the diagram. Write down what enters each stage, what output or evidence it produces, and which identity is allowed to perform it. Then test an empty input, an unauthorized request, a dependency timeout, and an unexpected result. For each case, state whether the application should retry, reject, abstain, or ask a person for help. Keep the observed failure as a regression example so changes to prompts, models, data, or code can be checked against the same behavior. Practice task: Why include trace and evaluation data in a privacy review?