Overview

The AI Usage Guidelines are organized into four categories, each controlling a different dimension of how you use AI at work. Every AI task you do is governed by all four at once. This page explains what each category covers and the single rule that governs when they point in different directions.

 

The Four Categories

Each answers a different question. You need to check all four before starting any AI task.


Governance Tiers

Do I need approval, and from whom?

Scope and risk level determine which tier you're in.

  • Tier 0 — Personal productivity, no approval needed
  • Tier 1 — Department-level use, supervisor should be aware
  • Tier 2 — Cross-department use, supervisor aware + CoE Lead and department head approve
  • Tier 3 — High-risk or autonomous actions, full CoE review. All AI agents are automatically Tier 3.

Data Classification

Can I use this data with AI at all?

The sensitivity of your data sets the outer boundary on everything else.

  • Level 1 Public — Approved enterprise tools permitted
  • Level 2 Internal — Approved enterprise tools permitted
  • Level 3 Confidential — Anonymize first, or use approved private environment
  • Level 4 Restricted — nebulaONE/TulaneAI private cloud models only (when available, with approval by CoE)

Approved Tools

Which platform am I allowed to use?

Tool restrictions apply at every tier and data level, no exceptions.

  • Approved — nebulaONE/TulaneAI, BoodleBox, Microsoft Copilot
  • Conditional — Browser extensions, SaaS AI features, custom AI skills (CoE review required before use)
  • Not approved for Tulane business — Personal ChatGPT, Claude, or any free LLM subscription. ChatGPT and Claude models are available through nebulaONE/TulaneAI and BoodleBox.

AI Use by Modality

Are there extra rules for this output type?

Output type rules stack on top of data and tool requirements.

  • Text — Human review required before use; label as AI-assisted if distributed externally
  • Code — Validate before deploying to any production system
  • Images — Label as AI-generated for any external audience
  • Video / Audio — CoE review required before any external use

▶ When the Categories Seem to Conflict

The most restrictive requirement always wins. Data classification restrictions apply regardless of your governance tier, the type of output you are creating, or what the quick reference card says. The categories stack — they do not override each other. A Tier 0 task involving Level 3 data still requires anonymization. A text output going externally still requires disclosure labeling even if the tier requires no approval.