What should colleagues know when AI helped prepare something they will use?
When AI helped prepare something your team will use, colleagues need to know how far to trust it. A short team agreement can say whether a comparison is preliminary, a source still needs checking or a draft isn't approved for external use.
AI can enter a shared assignment through one person's workflow before the team has talked about it. That's how mismatched assumptions about what's been checked get started.
Say which work the agreement covers, and which tools and uses of information are permitted under your organization's rules. The agreement gives nobody authority to override those rules.
I'd keep the agreement about responsibility. Naming the tool on a document tells the recipient nothing about whether the facts are current or whether anyone reviewed the result.
What you're after is a shared expectation about readiness. Colleagues shouldn't have to reverse engineer how a document was prepared to find out which parts they can rely on and which decisions are still open.
What to try this week
Try the agreement on one real assignment, then ask your colleagues which parts were unclear and revise those.