Requests that tripped a security-policy rule — e.g. trying to have the AI log in with credentials, hardcode a secret, or bypass a control. The AI still declined; this is for coaching/visibility.

Crawls the W-Training source-of-truth site (manager-only & system libraries excluded) into Azure AI Search. Runs in the background.

👍 / 👎 users gave on answers. Use the 👎 notes to fix prompts, SOPs, or the index.

Activity by user (click to expand): budget changes, re-index, email drafts, and 👍/👎 feedback. Append-only. To keep costs low we don't log every message — only these events.