Decision Flow
Work through these five questions in order for every AI task. All five apply every time — you cannot skip one because another looks straightforward. The most restrictive answer from any step governs what you can do.
Step 1
Classify Your Data
Identify the highest sensitivity level of any data you plan to use. This is the gating decision. Everything else depends on it.
Published content, marketing, public website
Process docs, internal memos, meeting notes
Financial data, FERPA records, employee data
SSN, HIPAA, payment cards — nebulaONE/TulaneAI private cloud only (check availability and approval with CoE)
Step 2
Select an Approved Tool
Match your data level to a tool authorized for that level. Do not select the tool first and assume the data fits.
nebulaONE/TulaneAI, BoodleBox, or Microsoft Copilot
nebulaONE/TulaneAI private cloud, or anonymize data first then use approved enterprise tools
nebulaONE/TulaneAI private cloud only (when available, with approval by CoE). No other AI tools permitted.
Step 3
Determine Your Governance Tier
Based on scope, risk, and whether an agent is involved. All AI agents automatically require Tier 3 regardless of data level.
Personal productivity, public or internal data only → no approval needed
Single department, some sensitivity → supervisor aware
Cross-department, shared workflows → supervisor aware + CoE Lead + department head approve
Data-modifying, agents, financial impact → full CoE review
Step 4
Check Modality Rules
What type of output are you creating? Modality rules add requirements on top of the tier. Both apply.
Internal: permitted with human review. External: label as AI-assisted if substantial.
Validate before production. External: additional review.
Internal: OK. External: must label as AI-generated.
Internal: OK. External: CoE review required + disclosure.
Step 5
Apply the Most Restrictive Rule
If your data level says "anonymize," your tier says "no approval needed," and your modality says "supervisor review for external use," every requirement still applies. You anonymize the data, and you get supervisor review before publishing externally. The tier telling you "no approval" does not cancel the other two. The categories stack. They do not override each other.
Example: You want to draft a summary of a facilities budget report using nebulaONE/TulaneAI.
Step 1: Budget figures are Level 3 Confidential → anonymize or aggregate before inputting.
Step 2: nebulaONE/TulaneAI private cloud is approved for Level 3 → tool selection is fine.
Step 3: Personal task → Tier 0, no approval needed from a scope standpoint.
Step 4: Text output for internal use → review before sharing; no external label required.
Step 5: The Level 3 data requirement is the most restrictive. Anonymize first, then proceed.