A project manager in a software firm's marketing team runs a Claude Project for campaign planning. A teammate wants to add the webinar sign-up list, which holds attendees' names, employers and email addresses, to the project knowledge. To stay within company policy, which bars customer personal data from AI tools, he proposes adding this to the project instructions: 'Do not repeat, store or share any personal data from the sign-up list.' Why does this proposal fail the policy?
- AThe instruction also needs repeating in every chat prompt, because project instructions are not applied to each chat.
- BThe instruction is too general, and it would comply if it named each personal data field Claude must leave out.
- CThe instruction shapes what Claude writes back, but the personal data has still been entered into the tool. Correct
- DThe instruction would comply if the list went into a single chat instead of the shared project knowledge.
Why A is wrong: This is tempting to anyone unsure how Projects work. It is wrong on two counts: project instructions apply across the Project's chats, and repeating the instruction would still leave the personal data entered.
Why B is wrong: This appeals because more precise prompts usually work better. It is wrong because no wording of an instruction undoes the fact that the data has been entered, which is what the policy prohibits.
Why C is correct: Correct. The policy controls what goes into AI tools, and adding the list to project knowledge puts the personal data in, whatever the instructions later tell Claude to do with it.
Why D is wrong: This is tempting because a single chat feels less exposed than a shared Project. It is wrong because the policy applies to any entry into an AI tool, so the location inside the tool does not change the outcome.