What Information Should You Never Share With an AI Assistant?
What not to share with AI: passwords, financial details, private records, client data, and other people's information. Safer ways to ask.
An AI assistant can help explain a problem or improve a draft without seeing every detail behind it. Before you paste text or upload a file, ask what the assistant actually needs to complete the task. The same check applies to screenshots, voice notes, spreadsheets, and connected files.
Method note: This practical guide was reviewed October 4, 2026 against the official sources linked below. The examples are fictional. It does not claim firsthand testing of an AI product; privacy settings and retention rules depend on the service and account.
Keep Passwords and Recovery Details Out
Do not enter passwords, passkeys, API keys, access tokens, recovery codes, one-time sign-in codes, private keys, or answers to security questions. Even a screenshot of a settings page can reveal one of these. If you need help with a login problem, describe the step that fails and the non-sensitive error message.
Instead of: “My password is [real password] and my recovery code is [real code]. Why can't I sign in?”
Try: “My sign-in fails after I enter a recovery code. What safe checks can I make without sharing the code?”
If you already entered a real secret, use the affected service's security settings to replace or revoke it and follow its account-recovery instructions. Our account-recovery readiness drill helps you prepare trusted recovery routes before an emergency.
Leave Financial and Identity Numbers Out
A credit card number, bank account or routing number, Social Security number, tax ID, driver's-license number, passport number, or full identity document does not belong in a general-purpose AI prompt. The same goes for a photo of a statement or tax form: it can expose names, addresses, balances, account details, and barcodes together.
Instead of: uploading your bank statement and asking what a fee means, describe the fee type and amount without the statement. If the question depends on the actual transaction or account, use the bank's official app or support channel. For tax or identity documents, use an authorized professional or service with the right handling rules.
Use General Questions for Medical and Legal Issues
Private medical records, lab results, prescriptions, insurance claims, legal correspondence, contracts, and case files may contain more about you and others than the question requires. A general question can often help you prepare for a conversation with a qualified professional.
Instead of: uploading a complete medical chart, ask, “What questions could I bring to my clinician about this type of test?” Instead of pasting a lawyer's confidential letter, ask, “What terms should I ask a lawyer to explain in a service agreement?” Do not use an AI answer as a diagnosis, treatment decision, or legal opinion.
Protect Company and Customer Information
Keep unreleased plans, customer lists, private meeting notes, pricing, source code, incident details, contracts, and internal URLs out of an unapproved assistant. A work account does not automatically mean every AI feature is approved for every company file. Check your employer's or client's rules and the exact product and workspace before submitting material.
Instead of: pasting a customer's complete complaint thread, describe the type of issue, the response you need, and the tone. For a real client workflow, use the client-data decision guide to check authorization and reduce the input first.
Ask Before Sharing Someone Else's Story
Other people's names, contact information, photos, addresses, medical details, job issues, private messages, or family situations are theirs to protect too. Changing a name to “Person A” may not be enough if the prompt still includes a rare job title, exact date, small town, or distinctive incident.
Instead of: “My colleague [full name] at [company] was disciplined on [date]; summarize these emails,” ask for a general, fictional template for documenting a workplace conversation. If the task truly requires the real details, use the organization's approved process.
Why Deleting the Chat Later Is Not a Safety Plan
Deleting a conversation can be useful after a mistake, but it happens after the service has already received the information. Removal from your visible history may not mean immediate removal from every system, and uploads or saved memories can have separate controls. Rules also vary by provider, product, and account.
For one concrete example, OpenAI's current ChatGPT retention guidance says deleted chats are removed from the account immediately and scheduled for deletion from its systems within 30 days, with stated exceptions; it also explains that some saved files are managed separately. The Memory guidance says deleting a chat does not automatically delete a separately saved memory. Check the current rules for the tool you actually use. The safer decision is made before you submit the information.
A Safer Way to Ask
Choose one of three approaches based on what the answer needs:
| If you need... | Share this instead | Keep outside the AI tool |
|---|---|---|
| Writing help | Purpose, audience, tone, and a short cleaned excerpt | Names, full email threads, account numbers, signatures |
| An explanation | A general question or invented example with the same structure | Private records and real customer data |
| Help with a real account or regulated matter | A non-sensitive description of the problem | Credentials, identity documents, medical or legal files; use an authorized channel for the actual details |
When an excerpt is necessary, remove names, emails, phone numbers, addresses, account and client identifiers, internal links, and access details. Generalize exact dates, amounts, and locations when they do not matter. Use placeholders such as [CLIENT] or [PROJECT] only when the roles need to remain distinct. Then read the complete prompt once more to see whether the remaining clues could still identify someone. The worked redaction example and checklist show this step in detail.
As a broad starting point, the UK's National Cyber Security Centre advises against putting sensitive information into public AI tools. The practical test is simple: if the task can be completed with a general description or fictional example, use that version.
If You Already Shared Something Sensitive
Stop repeating the information in follow-up prompts. For a password, token, or recovery code, change or revoke it through the affected service. For financial or identity data, contact the relevant institution through its official channel. For work or another person's data, follow the organization's reporting process. Use the tool's deletion and privacy controls where available, but do not rely on them as the only response.
Very clear and concise information.
ReplyDeleteOnline security is important when accessing cryptocurrency platforms. Users should be aware of Kraken phishing sites and fake login pages that may attempt to steal personal information or account credentials. Learning how to recognize suspicious links, verify official website addresses, and avoid fraudulent login pages can help protect accounts and maintain safer online activity. Want to know more? Visit our website for complete details: Кракен фишинг сайт