Lesson 1: What to Never Share with AI

In April 2023, Samsung employees accidentally leaked sensitive semiconductor designs to ChatGPT. They were trying to fix bugs in their code and improve test sequences. Three separate engineers pasted proprietary information into the chatbot within weeks of each other.

Samsung's response? They banned ChatGPT company-wide and limited prompts to 1,024 characters.

Here's the thing: those engineers weren't being reckless. They were trying to do their jobs faster. They just didn't realize that everything you tell ChatGPT can potentially be used to train future versions of the model or could be accessed in a data breach.

The fundamental rule for AI safety is simple: if you wouldn't post it on a billboard, don't put it in an AI tool.

The Four Categories of Dangerous Data

1. Personal Identifying Information

Don't paste your Social Security number, driver's license, passport details, medical records, financial account numbers, or your kids' school information into AI. Ever.

This seems obvious until you're trying to fill out a form or write an email about your health insurance. The temptation to ask AI for help with boring paperwork is real. Resist it.

Even partial information can be risky. Your full name, birthday, and city might seem harmless alone, but together they're often enough to find someone's identity.

2. Client and Customer Data

You know that NDA you signed? AI doesn't.

If you're a lawyer, don't paste case details. If you're in healthcare, don't describe patient situations (even without names). If you're a consultant, don't share client strategy documents. If you're in sales, don't upload your customer list to "help organize contacts."

In May 2023, a lawyer used ChatGPT to research cases for a court filing. The AI invented six fake cases, complete with fake citations. The lawyer didn't verify them. The judge was not amused. The lawyer got sanctioned.

The risk isn't just that AI will make things up. It's that your client's confidential information could end up in a training dataset, a data breach, or even just in OpenAI's logs.

3. Passwords, API Keys, and Access Tokens

Never, ever paste:

• Passwords or password reset links
• API keys or access tokens
• Security questions and answers
• Two-factor authentication codes
• Private keys or certificates

Don't ask AI to "generate a secure password and remember it for me." Don't paste error messages that include authentication tokens. Don't screenshot your settings page if it shows API credentials.

If you're debugging code and need to share an error message, scrub any keys or tokens first. Replace them with placeholder text like "YOUR_API_KEY_HERE."

4. Company Secrets and Proprietary Work

This is where things get tricky, because it's where AI is most useful.

You can't paste:

• Unreleased product details
• Financial data that isn't public
• Strategic plans or internal memos
• Proprietary code (unless your company explicitly allows it)
• Customer lists or sales data
• Anything covered by an NDA

But you can anonymize and generalize. Instead of "Here's our Q4 revenue projection for the European market," try "How should I structure a business presentation about quarterly forecasts?"

How Different AI Tools Handle Your Data

Not all AI tools treat your data the same way.

ChatGPT (OpenAI):

• Free version: your conversations can be used for training unless you opt out
• Plus/Team/Enterprise: you can disable training, but data is still stored for 30 days
• API: not used for training by default

Claude (Anthropic):

• Free and Pro: not used for training
• Conversations stored but not used to improve models
• Enterprise version offers additional controls

Google Gemini:

• Your conversations may be reviewed by humans
• Can be used to improve products
• Google Workspace version has stronger privacy protections

Microsoft Copilot:

• Business version: your data stays in your tenant
• Free version: covered by Microsoft's consumer privacy policy
• Enterprise version offers the strongest protections

The pattern here: free versions have weaker privacy protections. If you're using AI for anything work-related, consider paying for a version with business-grade privacy policies.

The Future-Proof Test

Before you paste anything into an AI, ask yourself: "What if this prompt became public tomorrow?"

Could it embarrass you? Get you fired? Violate a contract? Expose someone else's private information?

AI tools are new. Their privacy policies change. Companies get acquired. Data breaches happen. Court orders force companies to turn over data.

The safest assumption is that anything you tell AI could potentially be seen by others, either now or in the future.

What If You Already Shared Something Sensitive?

If you accidentally pasted confidential information into an AI tool, here's what to do:

1. Delete the conversation immediately. Most AI tools let you delete chat history. Do it now.

1. Change any exposed passwords or keys. If you pasted a password or API key, consider it compromised. Change it immediately.

1. Tell the relevant people. If you exposed client data, tell your legal team or compliance officer. If you exposed company data, tell your manager or IT security. Yes, it's uncomfortable. It's also necessary.

1. Request data deletion if possible. Some AI tools offer data deletion requests. ChatGPT has a form for this. Use it.

1. Document what happened. Write down what you shared, when, and with which tool. If there's ever a breach or investigation, you'll need this information.

The mistake isn't career-ending if you handle it responsibly. Covering it up and hoping nothing happens? That's career-ending.

Quick Reference Checklist

Before pasting anything into AI, verify it doesn't contain:

• [ ] Real names (yours or others)
• [ ] Contact information (email, phone, address)
• [ ] Financial account numbers
• [ ] Medical information
• [ ] Passwords or access credentials
• [ ] Client or customer data
• [ ] Proprietary company information
• [ ] Anything covered by NDA
• [ ] Information about minors

If you checked any of these boxes, don't paste it. Rewrite your prompt with anonymized, generalized information instead.

The Bottom Line

AI tools are incredibly useful. They're also incredibly bad at keeping secrets, because that's not what they were designed to do.

Use them freely for general knowledge, brainstorming, learning, and creative work. But draw a hard line at sensitive data. Your job, your clients, and your personal security depend on it.

When in doubt, leave it out.