The blocker on the rollout was never learner enthusiasm. It was a compliance team that could not approve a tool whose inputs and outputs had no defined handling rules — and was entirely right not to.
The framework replaced a general anxiety with a specific, answerable set of questions. Once those had written answers, approval became a decision rather than a risk.
| Tier | Definition | Assistant handling |
|---|---|---|
| Open | Already public or intended for publication | Full use. Encouraged as the practice surface for new users. |
| Internal | Ordinary business content, no regulatory or personal sensitivity | Full use with human review before anything is published or acted on. |
| Restricted | Personal data, commercial terms, security-relevant material | No input to the assistant. Named alternative workflows provided. |
| Regulated | Content under external regulatory obligation | Excluded entirely. No exception path at the individual level. |
The Author
Accountable for the output regardless of how it was drafted. The assistant is never the author — stating this plainly removed most of the ambiguity.
The Content Owner
Accountable for classification of source material and for keeping it current. Same ownership model as the LMS catalog, deliberately.
The Governance Review
Scheduled, not triggered. Reviews classification decisions and edge cases surfaced by users, and updates the framework rather than issuing exceptions.
- Written before the training, so guardrails were a design input rather than a retrofit
- Framed as four questions a person can answer at their desk in seconds, not a policy document to consult
- Reused the ownership and review-cadence model already operating in the LMS — familiar mechanics, new domain
- Became the prerequisite content that gated license provisioning, which is what made compliance the unlock rather than the obstacle