What NOT to Put Into AI Tools
AI Privacy Rule
Keep sensitive information out of general AI prompts, including names, family details, email addresses, phone numbers, account data, customer records, employee files, financial records, legal documents, medical information, and confidential business details. Use placeholders, redacted examples, or approved systems when needed, and keep human review before important actions. AI Privacy Rules
AI tools are useful for drafting, summarizing, organizing, and brainstorming, but they are not the right place for sensitive information. A safe AI workflow starts before the prompt is written. It starts with deciding what information should stay out of the tool entirely.
The basic rule is simple: if the information could identify a person, expose an account, reveal confidential business details, or create harm if shared with the wrong system, do not paste it into a general AI tool.
Personal Information
Do not paste full names, family names, home addresses, personal email addresses, phone numbers, birth dates, government IDs, school records, medical information, private photos, or location details into general AI prompts. Even when a task seems harmless, personal details can create privacy risk.
Use placeholders instead. Replace real names with labels like [customer], [employee], [student], [family member], or [client].
Account and Security Information
Never paste passwords, API keys, access tokens, login links, private URLs, authentication codes, recovery phrases, account numbers, payment details, or security configurations into AI tools. These details can create direct account or system risk.
If you need help troubleshooting a technical issue, describe the problem without sharing credentials or live access details.
Customer, Employee, and Business Records
Customer records, support tickets, CRM notes, employee files, candidate evaluations, payroll details, contracts, vendor records, pricing strategy, and internal documents should be handled carefully. Use approved systems when your organization requires them, and remove sensitive details before asking AI to summarize or draft anything.
Legal, Financial, Medical, and Regulated Information
Be especially careful with legal documents, financial records, tax documents, medical notes, insurance information, compliance records, and hiring or employment materials. AI can help create a checklist or organize questions, but it should not replace qualified professional review.
Use Redacted Examples
Most AI tasks do not need real private data. You can ask for a draft, checklist, summary format, or response structure using a redacted example. This keeps the workflow useful while reducing exposure.
For example, instead of pasting a customer’s full complaint with identifying details, write: “A customer says their order arrived late and wants a refund. Draft a polite response structure for human review.”
What to Do Instead
- Use placeholders for names, accounts, locations, amounts, and dates.
- Summarize sensitive situations instead of pasting raw records.
- Use approved tools for private or regulated workflows.
- Keep human review before sending, publishing, deciding, or automating.
- Delete unnecessary sensitive details before saving or sharing outputs.
AI works best when the information is limited, the output is reviewable, and people remain responsible for final action. Protect the input first, then verify the output before using it.
Example in Practice: Redacting a Payroll Question
Don’t paste this: “Maria Gonzalez (employee #4471, salary $68,500) was overpaid $312 on the June 1 run because her new deduction wasn’t applied. Draft an email explaining the correction.”
Paste this instead: “An employee was overpaid a small amount on the last payroll run because a new deduction was missed. Draft a short, respectful email explaining the error, the correction on the next run, and who to contact with questions.”
Result: The same usable draft. The name, employee number, salary, and amount are added back inside your own payroll system — not in an AI tool’s history.
The habit: Write the prompt as if it could be read aloud in the break room. If it couldn’t, redact more.
Sources & Further Reading
- OWASP Top 10 for LLM Applications — see LLM02: Sensitive Information Disclosure.
- FTC: Artificial Intelligence guidance and cases — how regulators treat mishandled consumer data in AI products.
Reviewed against the 4AIWorld editorial approach · Updated June 2026