Code of Conduct client

Lilly

Controlled — Impact-Based Approval

Lilly frames AI use around impact, not tool: a Low-Impact use — AI as a pure productivity aid, or a no-retention pass that never trains on Lilly content — can proceed without sign-off. Anything else, including any use that touches Lilly content and gets stored or reused, needs Lilly's prior written approval.

Key requirement — Lilly's system is Low-Impact vs. High-Impact Use, not a generic tier scale. Low-Impact only covers two narrow cases: a productivity tool that never touches Lilly content or feeds a deliverable, or a no-retention tool that processes Lilly content without storing or training on it. Everything else needs Lilly's prior written approval before you start.

Permitted

Low-Impact Use: an AI tool used purely as a productivity aid, with no Lilly content going in and no AI output going into a deliverable
Low-Impact Use: Lilly content processed by a no-retention tool that doesn't store inputs/outputs or use them as training data
Anything beyond that, once Lilly has given prior written approval

Not permitted

Using any AI system with Lilly content, or in connection with Lilly services, without Lilly's prior written approval (unless it qualifies as Low-Impact Use)
Training any AI model on Lilly content
Treating an ambiguous case as Low-Impact by default — if in doubt, it's High-Impact

Tool-by-tool approval

LynxAI

High-Impact Use needs Lilly's prior written approval

Conditional

AVA

OK as Low-Impact Use if no Lilly content is stored/retained

Conditional

Microsoft Copilot

OK as Low-Impact Use if no Lilly content is stored/retained

Conditional

NotebookLM

OK as Low-Impact Use if no Lilly content is stored/retained

Conditional

Vermeer