TL;DR - Never paste secrets into a public AI tool. Run inputs through a quick safe-use checklist and redact anything you couldn't post publicly.
Why it matters
The fastest way to get into real trouble with AI isn't a wrong answer - it's leaking data. Consumer tools may use what you type to improve their models, and one pasted contract or customer list can become a serious problem. The good news: a 10-second habit prevents almost all of it.
Never paste into a public AI tool
- Customer or employee personal data (names, emails, IDs).
- Trade secrets, unreleased financials, source code you don't own the rights to share.
- Passwords, API keys, anything confidential.
- Basically: anything you wouldn't post on a public forum.
Worked example - redact, don't expose
Risky: paste a real contract: "Acme Corp pays Jane Doe $8,500/month..."
Safe: "Company A pays Person B $X per month..." - you still get help with the structure and risks, without exposing the deal.
Steal this - the safe-use checklist
1. Would I be okay if this input were public? If not -> redact.
2. Is there an approved/private AI tool at work? Use it for sensitive stuff.
3. Did I read the OUTPUT before sharing it? (You own it.)Common mistakes
- Pasting the whole document when you only needed help with the format.
- Assuming "private chat" means private. Free tools often train on inputs.
- Shipping AI output unread - including a hallucinated fact or wrong tone.
Good to know
Many tools have a safer enterprise/business tier: ChatGPT Enterprise/Team, Claude for Work, and Gemini for Workspace typically do not train on your data and add admin controls. If your company has one, use it for work content. On free consumer tiers, check the data settings - some let you turn off training.