---
name: adopt-ai-concern-statement
description: Puts one person's concern about how AI is used at work into their own words as a four-part statement, with the question to ask a specialist and a choice of route including a private one. Use for "run adopt-ai-concern-statement", "help me raise a concern about AI", "how do I say this without sounding against AI", "word my concern about the AI rollout", "I see a problem with how we use AI", "write up my AI concern", "who do I raise an AI concern with", part of the Claude Guide for AI Adoption Pack by Polar Bear.
---

# AI Concern Statement

## When To Use
You can see a problem with how AI is being used at work, and you cannot say it without sounding like you are against progress. Use this when the concern is real, the words are not there yet, and you have not decided whether to raise it. It answers: what exactly am I saying, what am I asking for, and through which route?

## When Not To Use
Do not start here with a serious allegation or a real fear of retaliation: go first to a trusted human channel, and come back for wording only if that channel asks for it. If the statement is written and you want to practise the conversation, run Speak-Up Practice Round. If you are the one receiving concerns, run Concern Response Log.

## Inputs
- Your account of what happened, as you would tell it to a colleague. Rough notes are enough.
- Who you are thinking of raising it with, and what you would like to happen.
- Any limit on what you can share, and whether raising it could put you at risk.
If you have none of this, I start from one sentence about what is bothering you and mark the output as a first draft.

## Approach
The method is a plain four-part structure with no owner: observation, interpretation, consequence, request. Keeping the four apart is the whole craft, because a concern that opens with a conclusion ("this tool is dangerous") gets argued with, and one that opens with what you saw gets looked at. The failure it prevents is the concern polished into an apology: three softening sentences, a compliment to the project, and a request so vague nobody has to do anything. The statement stays yours. AI help with the wording proves nothing about whether the concern is right.

## Workflow
1. Ask up to three questions, only where your account leaves a gap: who it is for, what you want to happen, and whether raising it could put you at risk. If the third answer is yes, the route comes before the wording.
2. Split your account into four parts and keep them apart: what I saw (one concrete example), what I think it means, what could follow, what I am asking for. With no concrete example, I frame a question instead of inventing evidence.
3. Keep the strongest legitimate point and say it straight. Guesses about anyone's motives come out. A real concern is not softened into an apology or a suggestion.
4. Where the issue crosses into someone else's field (legal, security, data), turn your conclusion into a question that specialist can answer. You do not need their expertise to ask it, and I do not lend you any.
5. Write a short opening in your voice and at your level of formality. If it sounds like a press release, it is wrong.
6. Lay out the routes: a direct conversation, a written note, an intermediary, an established confidential channel. For each, what it asks of you and what record it leaves. Nobody is pushed to speak in public or to confront someone they fear, and no route is described as anonymous: workplace tools keep records. Then the statement goes back to you: you decide whether, when and with whom to share it, and I send nothing.

## Output Format
```markdown
# AI Concern Statement
## What I saw
[One concrete example: the task, what happened, when. No guess about why.]
## What I think it means
[My reading of it, stated as my reading. What I know and what I do not.]
## What could follow
[The consequence for the work, the customer or the team if nothing changes.]
## What I am asking for
[One request someone can act on: a check, a pause, a rule made clear, a conversation.]
## Questions for a specialist
| What I am not placed to judge | Question to ask | Who could answer |
|---|---|---|
| [legal, security, data or other point] | [question] | [role; legal points: check with a qualified adviser] |

## My opening
[Two or three sentences in my own voice.]
## Routes
| Route | What it asks of me | What record it leaves | Private |
|---|---|---|---|
| [direct conversation / written note / intermediary / confidential channel] | [effort and exposure] | [what is kept, and where] | [yes or no] |

## Decision
I decide whether to share this, with whom and by which route, by [date I choose]. Not sharing it is one of the options.
```

## Done When
- The four parts are separate, and "what I saw" holds a concrete example or an honest question.
- The request is one thing a named role could do, and every point outside your own expertise is a question, not a claim.
- At least one route is private, no route is described as anonymous, and the decision is left to you with no pressure attached.

## Quality Bar
- The words are yours: your phrases, your level of formality, your degree of certainty.
- Nothing is invented (no audit, no finding, no "the AI confirmed it"), and no comment is made on your courage, your credibility or your attitude to AI.
- Legal protection for people who raise concerns is not explained here: check with a qualified adviser.
- The person decides whether, when and with whom to share it; Claude drafts and sends nothing.

## Next
Run adopt-speak-up-practice (Speak-Up Practice Round) to rehearse saying it before the real conversation.

## About the makers

This pack is made by Polar Bear, a consultancy built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).