Protect Privacy, Money, Identity, and Client Data

AI tools invite people to paste information into a conversation. A note, email, invoice, medical summary, or client record may feel like ordinary text. Once submitted, it can be stored, reviewed, used to improve a service, or exposed through an account takeover. The privacy question starts before the first prompt.

This chapter uses a simple rule: enter the minimum information needed, only into an approved service, for a clear purpose. If the purpose cannot be explained plainly, or the service cannot support the required protection, use a manual route.

Classify information before sharing

Divide information into four categories:

• Public: Already intentionally published and safe to repeat.
• Personal but low risk: General preferences, public interests, or non-sensitive draft ideas.
• Confidential: Private correspondence, business plans, unpublished creative work, or personal details.
• Highly restricted: Passwords, access codes, government identifiers, bank details, health records, precise location, and private data about children or vulnerable adults.

The category should guide the route. An approved organizational system may permit confidential work under policy. A free public chatbot may not. Highly restricted information should not be entered merely because the tool offers a privacy setting.

Never send credentials or recovery codes

No legitimate support person needs your password, one-time verification code, recovery phrase, or remote access without a clear, independently verified process. AI services can be impersonated. A convincing chat window is not proof of identity.

Use separate credentials for important accounts, a password manager where appropriate, and second verification. If someone contacts you unexpectedly and asks for access, end the contact and use an official number or website you find independently.

Get consent for other people's data

You may be allowed to see a client's information without having permission to place it in an AI service. Employment, volunteer, family, and business relationships can create duties of confidence.

Ask:

• Who owns or controls this information?
• Was consent obtained for this use?
• Does the organization permit AI processing?
• What data does the service retain or use?
• Can the task be completed with a redacted example?
• Who can access the output?
• How will the data be deleted?

If these questions cannot be answered, do not paste the material. Use fictional or redacted examples for drafting.

Separate money from assistance

AI can help explain a bill or draft a question, but it should not receive bank passwords or authorize payments. It can misread a document or be manipulated by hidden instructions. An account holder remains responsible for checking a payment, recipient, amount, and terms through the official banking route.

Be careful with:

• open documents or screenshots that include account numbers
• shared cloud files with links that still work after the task
• auto-fill on a shared device
• a browser extension that reads every page
• saved conversation history containing financial details
• emails or messages directing you to an unverified payment link

Use a trusted adviser or the responsible institution for a consequential money question. Do not treat a generative answer as financial authorization.

Identity documents require a specific route

Government identification, tax records, medical cards, immigration papers, and proof of address contain information that can enable fraud. Before sending a copy, confirm:

• why the recipient needs it
• whether a less sensitive alternative exists
• how it will be transmitted
• who will access it
• when it will be deleted

A photograph may remain in a phone's gallery, cloud backup, message history, and AI service history. Deleting it from one place does not necessarily remove every copy.

Protect client and community data

If you help a small business, nonprofit, or community group, use organization-owned accounts and approved storage. Do not copy records into a personal AI account to save a few minutes. Do not put names, case notes, schedules, health details, or children's information into an unapproved tool.

For AI-assisted administrative work, a practical split is:

• keep raw records in the approved system
• remove identifiers before asking for formatting help
• review the output against the source
• return the final result to the approved system
• delete temporary copies

This workflow still requires the organization's permission. It is not a guarantee of legal compliance.

Check retention and access

A service may keep conversations for product improvement, moderation, or account history. Settings can differ by country, account type, and product. Read the current terms rather than assuming that closing a window deletes the data.

Ask:

• Does the service retain the conversation?
• Can a human reviewer access it?
• Can it be used for training?
• Is deletion immediate or delayed?
• Does the backup or device keep another copy?
• Who else has access to the account?

Record the answer if you are using the tool for paid or organizational work.

AI cannot give informed consent

An AI may claim that a use is safe, lawful, compliant, or private. That is output, not consent, approval, or legal advice. The responsible organization or professional must establish those facts.

This matters in health, legal, financial, employment, housing, and care settings. It also matters in ordinary participant communication. A model can draft a consent paragraph, but a person must ensure that it accurately describes the use and gives people a real choice.

Exercise: Redact one document

Choose a non-critical document you would never share unredacted. Make a copy and remove:

• names and contact details
• account or file identifiers
• addresses and precise dates of birth
• health or care details
• confidential business figures
• any instructions that grant access

Replace them with consistent labels such as [CLIENT], [DATE], or [AMOUNT]. Ask whether a generic draft can be produced from the redacted version. If not, the task requires an approved source and a human reviewer rather than a public chatbot.

A fictional privacy failure

A fictional volunteer coordinator wants help summarizing feedback forms. She photographs every form and uploads the images to a free chat service because the service recognizes handwriting well. One form includes a participant's medication and a phone number.

The safer path begins before the upload: the organization's policy determines whether AI may process the data, the forms are stored in the approved system, and a de-identified summary is created. If the service is not approved for personal data, staff summarize on paper or use approved software. Convenience does not create authority.

Responding to a suspected exposure

If private data was entered into the wrong tool or account:

1. Stop using the conversation.
2. Preserve relevant facts without copying more private data.
3. Follow the organization's incident or privacy process.
4. Change affected credentials and revoke unnecessary access.
5. Ask the service and responsible adviser about deletion or notification.
6. Warn affected people if the law or policy requires it.

Do not hide the mistake because the tool produced a polished result.

What to carry forward

AI privacy is a design choice repeated at every paste, upload, and account login. Use the minimum data, verify the service, get permission for other people's information, and retain an offline route. The next chapter puts privacy, verification, and access into one small human-checked workflow.