The first time someone asked me to help them use AI at work, the very first thing they did was paste a spreadsheet with names, ID numbers and health data from their patients. I froze. Not because the AI was going to “steal” that data, but because that person had no idea where it was going.
And the truth is, most people who are new to AI don’t know either. That’s why I’m writing this.
Where your data goes when you use an AI
When you use a tool like ChatGPT, Claude or Gemini, the messages you write travel to that company’s servers. There, the AI processes your text and sends back a response. That’s cloud AI: the AI runs on the provider’s computers, not on yours.
What happens to that text afterward depends on each company’s policies. In general:
- Many offer a mode where your conversation is not used for training (with Claude, for example, paid conversations are not used to train the model by default)
- Some store conversations temporarily to improve the service
- On free plans, the terms are usually less strict, so it’s worth reading the fine print
This doesn’t mean AI is a spy. But it does mean you shouldn’t send just anything without understanding what happens to it.
What you should NEVER share with an AI
There is information that, regardless of the platform, should not leave your infrastructure:
- Personal data of your clients or patients: full name, ID number, account number, medical history, address
- Passwords or access tokens: never, under any circumstances
- Confidential financial information: account statements, card numbers, employee payroll data
- Trade secrets or intellectual property: formulas, proprietary code, business strategies you don’t want leaked
- Legally protected information: privileged communications (like attorney-client) or data regulated under healthcare or banking laws
If you need to work with this type of information, consider tools with enterprise plans (which include confidentiality agreements) or open-source models that run on your own server.
What IS safe to share
You can use AI confidently for:
- Drafting texts, emails and proposals (with general information, no third-party personal data)
- Summarizing public documents or ones without sensitive information
- Generating ideas, content structures or strategies
- Analyzing already anonymized data (no names or identifiers)
- Learning, studying and asking questions about any topic
The simple rule: if you’d read it aloud in a meeting without a second thought, it’s probably fine to pass it to an AI.
How to protect yourself without overcomplicating things
You don’t need to become a cybersecurity expert. These basic practices cover most of the risks:
1. Read the privacy section (even just a quick scan) Find the part that explains how they use your conversations. Claude and ChatGPT have clear documentation on this; it takes five minutes tops.
2. Turn on the available privacy options In ChatGPT you can turn off conversation history. In Claude, paid plans don’t use your data for training by default. Check what options your tool has and enable them.
3. Anonymize before pasting If you need to analyze client data, remove or replace names and identifiers before sending them. “Client A” instead of the real name. “City X” instead of the address. The AI can still help you without knowing who the person is.
4. Use enterprise versions for sensitive work If your company handles regulated information, enterprise versions (Claude for Work, ChatGPT Enterprise) come with stronger privacy agreements. Talk to your tech or legal team before deciding.
5. Trust your gut If something feels like you shouldn’t send it, you probably shouldn’t. Your judgment is a solid first line of defense.
AI isn’t the problem, lack of judgment is
Artificial intelligence is a powerful tool, and using it well means understanding its limits, including privacy ones. It’s not about being afraid of it; it’s about using it thoughtfully.
Once you learn to separate what’s safe to share from what isn’t, you can take full advantage of everything AI has to offer without putting your clients, your company or your data at risk.
Next time you’re about to paste something into an AI, take one second: is there information here I wouldn’t want someone else to see? If there is, anonymize it first. It’s a habit that takes five seconds and can save you enormous headaches down the road.
Want these tools compared in depth? Check the unbiased reviews.
