AI Policy

As a lab, we recognize the potential of AI tools to advance, clarify, and accelerate our work and learning. These benefits do not come without substantial social, environmental, and technical costs. As a result, we have adopted the following guidelines for our use of AI tools as part of our lab activities:

Lab members are responsible for all content they create, generated by AI or otherwise.
If you don’t feel like you can or want to create content without AI, that’s okay! You can say that. It’s better to admit that something doesn’t feel possible than to rely on AI, but it’s also okay to use AI if it feels helpful rather than necessary.
All content produced by lab members for internal or external use should include an AI Attribution.
1. You can generate a basic AI attribution using https://aiattribution.github.io/
2. All content is assumed to be “AIA No AI v1.0” (entirely human generated) unless otherwise attributed. All content, even AIA No AI v1.0, should be attributed when possible.
3. Any content used in settings where a full AI Attribution is impractical (for example, in a short email) should not use AI.
Prompts, AI tools, and dates used to generate or modify any public facing content must be recorded and shared as part of the content that they generated.
1. This includes images, text, and code.
2. For example, include in the notes of a powerpoint slide what prompt was used for the image that generated it, which AI system, and the date that AI was used, along with the attribution (”: 20260717 AIA EAI Hin R Midjourney v1.0” for a completely Midjourney-created image on July 17, 2026, for example).
3. If content was produced interactively, then key aspects of prompting can be recorded. Any sufficiently complex or detailed prompting history that was integral to content creation must be saved directly.
“Indirect” uses of AI, such as to learn material for a given project; search for literature; refine ideas that then result in the given content; act as a code completion engine; or otherwise must be recorded for all related downstream content produced.
1. Content can have multiple AI Attributions.
2. IAIA No AI v1.0 can been used when there is no indirect AI use.
3. (Separation of creative vs technical use of AI — web searches that integrate AI — etc)
“Negative” uses of AI (finding content you DON’T want to include using AI) should be attributed as AIA Anti-AI v1.0 .
Ultimately, the decision about what is a “direct” or “indirect” use for AI is a personal one made by everyone in lab. When in doubt, we will share more about any AI use in our work.
Do not publicly share AI generated content attributed to the lab.
1. This includes text, images, and data.

What content can go into AI tools?

Only Fred Hutch-approved uses of AI are acceptable in our lab. This means only using Perplexity for proprietary information, not putting any individual PHI or PII into AI systems, and setting up a Data House Call to confirm any AI policy concerns before using AI tools.
1. If there are tools you would like to use and DaSL does not support their use, then please let the lab know at a lab meeting or let Nasa know, and we can discuss and decide how to advocate for updating our institutional policy.
Do not put whole documents other people wrote into AI without their permission, whether for interpretation, summarization, validation, or otherwise.
Do not put whole codebases that are not public into AI.
1. Commit your code to a public repo and share the link if using this approach.

TO ADD: USE CASES FOR AI. SEE https://rilab.ucdavis.edu/ai_expectations.html

How can we set up review processes for AI?

All AI generated content must be reviewed by at least two lab members before being made public or shared.
1. This includes both direct and indirect use.
2. This includes both internal and external content.
All AI generated code must be reviewed by at least two lab members before being run on real data.