How AI Tools Can Put Your Data at Risk, and How to Use Them Safely
Artificial intelligence tools such as ChatGPT, Gemini, Copilot, Claude, and AI features built into browsers and productivity software can save enormous amounts of time. They can summarize documents, analyze spreadsheets, write content, create images, research topics, and help solve technical problems.
But there is an important tradeoff: AI cannot work with information you provide without processing that information somewhere.
That does not mean AI companies are literally “stealing” everything you enter. The more accurate concern is that users frequently give AI systems far more information than they realize, sometimes under privacy settings they have never reviewed.
What AI Tools May Collect
When you type a prompt into an AI service, you are providing data to the company operating that service. Depending on the product, account type, and settings, that information may include:
- Your prompts and conversations
- Documents and spreadsheets you upload
- Photographs and other images
- Audio or voice recordings
- Feedback you provide
- Device and usage information
- Approximate location information
- Information retrieved from connected services
- Email, calendar, cloud-storage, or other information when you explicitly connect those services
For example, Google’s current Gemini privacy documentation says information provided to Gemini can include prompts, files, videos, screens, photos, imported chats, and browser page content. Google also says a subset of data can be reviewed by trained human reviewers. (Google Support)
This is why privacy settings matter.
The Biggest Mistake: Treating AI Like a Private Notebook
People naturally begin talking to an AI assistant conversationally. That makes it surprisingly easy to paste information that should never have been entered.
A useful rule is:
Don’t give an AI service information you wouldn’t be comfortable storing on someone else’s computer.
That is deliberately conservative, but it is an excellent habit.
For example, avoid entering passwords, authentication codes, Social Security numbers, complete financial account information, private medical records, confidential business information, unpublished customer databases, private personnel information, or other people’s sensitive personal information.
Instead of entering:
“Here is my customer’s complete account record. Analyze it.”
remove identifying information first:
“Here is an anonymized customer record. Analyze the purchasing pattern.”
You usually get essentially the same AI benefit with substantially less privacy exposure.
Your Conversations May Be Used to Improve AI
This is where the details become important because different AI services have different policies, and different versions of the same service may have different policies.
For example, OpenAI says conversations from individual ChatGPT services may be used to improve models unless the user opts out. ChatGPT users can currently go to Settings > Data Controls > Improve the model for everyone and turn the setting off. Future conversations will remain in chat history but will not be used for model training. (OpenAI Help Center)
ChatGPT also provides Temporary Chat. Those conversations do not appear in history, do not create memories, and are not used to train OpenAI’s models. OpenAI says Temporary Chats are deleted from its systems after 30 days, although they may be reviewed for abuse monitoring. (OpenAI Help Center)
Google provides its own controls for Gemini. Its documentation says users can manage Gemini Apps Activity and use temporary chats. Google also cautions users not to enter confidential information they would not want a reviewer to see or Google to use for improving its services. (Google Support)
Microsoft’s policies illustrate another important distinction. Consumer Copilot users can control whether conversation activity is used for model training, while Microsoft says prompts, responses, and file contents used with Copilot inside Microsoft 365 Personal and Family apps are not used to train foundation models. Work and enterprise protections can differ again. (Microsoft Support)
The lesson is simple: don’t assume every AI tool, subscription, or account has the same privacy rules.
Be Especially Careful With Connected AI
Modern AI assistants can connect to email, calendars, documents, cloud storage, and other applications.
This makes them dramatically more useful. It also increases the amount of information they potentially have permission to process.
Before connecting an AI assistant to another service, ask three questions:
- What information am I allowing it to access?
- Does it need that information to perform the task?
- Can I revoke that access later?
Use the principle of minimum necessary access. If an AI tool only needs access to one service, don’t give it permission to five.
Uploading Documents Creates Another Risk
Suppose you want AI to summarize a 40-page report.
Before uploading it, check whether the document contains information the AI doesn’t actually need.
Remove names, addresses, account numbers, employee information, confidential financial figures, hidden spreadsheet columns, document comments, revision histories, or other sensitive material whenever practical.
The goal isn’t to stop using AI.
The goal is to provide the minimum amount of data necessary to complete the job.
Free AI Tools Deserve Extra Scrutiny
A recognizable AI model name does not necessarily tell you who receives your information.
A website might advertise itself as an “AI PDF Analyzer,” “AI Resume Writer,” or “Free ChatGPT Tool.” Your document may actually pass through the website operator’s servers before reaching an AI provider.
That introduces another company into the data chain.
Before uploading sensitive material to an unfamiliar AI website, determine:
- Who operates the service?
- Is there a privacy policy?
- How long is information retained?
- Is uploaded content used for AI training?
- Is information shared with third parties?
- Can you delete your information?
- Is the tool actually operated by the AI company you think you’re using?
If those answers aren’t clear, don’t upload confidential information.
A Simple AI Privacy Checklist
Before pressing Send, take five seconds and ask:
Would this prompt expose me, another person, my organization, or a customer if it were seen by someone else?
Then follow these habits:
- Remove personally identifiable information whenever possible.
- Never enter passwords or authentication codes.
- Don’t paste complete financial or identity records.
- Review an AI service’s training and privacy controls.
- Use temporary or private conversation modes when appropriate.
- Remove unnecessary information before uploading documents.
- Review permissions for connected applications.
- Periodically delete conversations you no longer need.
- Be particularly cautious with unknown “free AI” websites.
- Use business or enterprise AI accounts for confidential organizational work when their contractual protections are appropriate.
AI Isn’t the Problem. Uninformed Sharing Is.
AI tools can be extraordinarily useful. Avoiding them entirely isn’t necessary for most people.
The better approach is to change how we think about the prompt box.
A prompt box is an information-sharing form.
Every time you type something, upload a photograph, attach a document, or connect another application, you’re making a decision about information access.
Before giving an AI more information, ask:
Does the AI actually need this information to accomplish what I want?
If the answer is no, leave it out.
That single habit can provide much of the productivity benefit of artificial intelligence while significantly reducing unnecessary exposure of your personal and business information.
