Privacy & Security

Is It Safe to Paste Client Data Into Your Agent?

A simple decision workflow for handling client/customer information without oversharing sensitive data.

⏱ About 4 minutes
Why this matters

A recurring community question is whether private client details are safe to paste into an agent. The right answer is: only the minimum needed, with redaction first.

The 60-Second Safety Rule

If you would not post it in a team channel, don't paste it raw into an AI prompt. Use anonymized versions first. Add real details only if absolutely required.

Fast Workflow (Use This Every Time)

1

Classify the data before you paste

Split your content into three buckets:

2

Redact first, then prompt

Replace personally identifying details with placeholders before sending.

Client Name → [CLIENT_A] Email → [CLIENT_EMAIL] Phone → [CLIENT_PHONE] Account Number → [ACCOUNT_ID]

Most planning, writing, analysis, and workflow tasks work fine with placeholders.

3

Use minimum necessary context

Don't dump full histories. Send only the small excerpt required for the task at hand. Less data = lower risk + lower token cost.

4

Add a privacy contract to your prompt

Include a short instruction block every time sensitive context is involved:

Treat all provided data as confidential. Do not repeat raw identifiers in your final answer. Use placeholders in outputs unless I explicitly ask for exact values. If more detail is needed, ask for one field at a time.
Success check

Your agent completes the task without exposing names, direct identifiers, or secrets in chat output.


When You Should Not Use AI for the Raw Data

Never paste these

Passwords, API keys, private tokens, MFA backup codes, or anything that can grant account/system access.


Copy/Paste Safe Prompt Starter

I need help with [task]. I will provide anonymized client data only. Rules: 1) Keep placeholders in your output. 2) Do not include personal identifiers. 3) Ask only for the minimum additional field if required. 4) Return a concise answer in bullet points.

FAQ

What if I already pasted sensitive info?

Stop and rotate any exposed credentials immediately. Then switch to redacted workflows and avoid reposting the same raw data in follow-ups.

Can I use real names if I trust my workspace?

Use placeholders by default. Real names should be the exception, not the norm, and only when required for the exact task.

How do I ask for help safely in #help?

Share issue behavior and exact error text, but remove client identifiers and secrets. You can include redacted examples instead.