A library GenAI roadmap has no plan to retire models trained on donor data that later becomes restricted. What long-term security planning gap exists?
Select an answer to reveal the explanation.
Short Explanation
When donor data becomes restricted, you need a retirement plan for models trained on it, not a forever leftover. Bookmark fonts, weekend movie recommendations, and Wi-Fi SSID branding updates don’t provide that offboarding control.
Full Explanation
The long-term planning gap is missing lifecycle offboarding and decommission controls for models trained on donor data that later becomes restricted. Your roadmap should inventory affected models and datasets, define legal or privacy triggers for restriction, withdraw serving endpoints, securely delete or isolate model weights and caches, purge retrieval indexes, and verify that restricted donor data no longer influences outputs. Without these controls, training that was once lawful can become a compliance and trust failure. Bookmark serif or sans-serif fonts concern presentation, not model retirement or restricted-data exposure. Weekend movie recommendations describe a chatbot capability, not a control for withdrawing a model or removing donor data. Routine Wi-Fi SSID branding updates may signal AI adoption, but branding does not decommission models or prevent residual data exposure. Exam caveat: removing the chat interface alone is not decommissioning; weights, embeddings, logs, and backups may still retain restricted donor signals. Practical check: add a model-and-data offboarding runbook to the roadmap, with restriction triggers, accountable owners, deletion or isolation steps, and audit evidence for changes in donor terms or law.