do not paste anything into an AI tool at work that names or identifies a real person, or that your employer has not approved for that tool. That means no names, no account or card numbers, no patient details and no internal figures. For a frontline worker, the risky moment is usually a copy and paste. The rule is short, but what counts as sensitive differs between a medical billing office, a bookkeeping desk and a customer support queue. The table below gives each role a never / ask first / fine list.
This is general information, not legal advice. Your employer's written policy and its privacy or compliance officer decide what applies to you. Federal guidance is at HHS's HIPAA pages, and Massachusetts rules are at 201 CMR 17.00.
Why this keeps happening
Workers often use AI tools their employer never chose. Microsoft and LinkedIn's 2024 Work Trend Index surveyed about 31,000 people in 31 countries. It found that "about 78 percent of users are bringing their own AI tools to work," and that only 39% had received training on them, as PCWorld reported in May 2024. That survey covered knowledge workers, and Microsoft sponsored it.
Frontline workers are often left out of the conversation altogether. In a Jobs for the Future survey of 3,020 people, fielded in late 2025, 56% of workers said their employer had never consulted them on how AI is used, per JFF's 2026 survey page.
The federal AI literacy framework treats this as a core skill. TEN 07-25.pdf), issued 13 February 2026, says workers "should understand what types of data should not be entered into AI tools" and "how to prevent accidental disclosure of confidential information." The framework is voluntary guidance for training programs, not a rule that binds your employer.
The three-bucket rule
Sort every piece of information before it goes near a prompt.
- Never: anything that identifies a real customer, patient or coworker, or gives access to money. Remove it, or do not use the tool.
- Ask first: your employer's internal information, and any real case details, even with names removed. Use it only in a tool your employer has approved in writing for that kind of data.
- Fine: public information, made-up examples, and your own words describing a situation in general terms.
The role-by-role table
ROLE: MEDICAL BILLING / FRONT DESK
NEVER Patient names, dates of birth, addresses, phone numbers, emails
Medical record numbers, health plan member or beneficiary numbers
Dates of service tied to a person, diagnoses or procedures
tied to a person, claim numbers, insurance card images
ASK FIRST Real claim scenarios with identifiers removed
Payer contract terms, internal fee schedules
Any tool your practice says has a business associate agreement
FINE Public payer policy text, published code descriptions
"Write a polite appeal letter template with placeholders"
Invented patients: "Pat Example, claim 0000"
ROLE: BOOKKEEPING / ACCOUNTS
NEVER Client or employee names with Social Security numbers,
driver's license numbers, bank account, card or routing numbers
Login credentials, PINs, passwords, payroll records
Tax returns or bank statements, in any form
ASK FIRST A client's real transactions with names and numbers removed
Your firm's internal financial reports or budgets
Uploading spreadsheets to any AI tool
FINE Public accounting rules and tax agency guidance
Invented ledgers: "Bakery LLC, 20 sample transactions"
"Explain the difference between an owner's draw and an expense"
ROLE: CUSTOMER SUPPORT
NEVER Customer names with order numbers, emails, phone numbers
or addresses; payment card details; account passwords
Screenshots of a customer's account page
Complaints or chat logs that identify a person
ASK FIRST Real tickets with all identifiers removed
Internal policies, refund rules, escalation procedures
Any AI feature built into your ticketing software
FINE Your company's public FAQ or published return policy
"Make this reply warmer and shorter" on text you wrote,
with no customer details in itThree legal anchors sit behind those rows.
Health information. For a practice covered by HIPAA, a vendor that handles protected health information on the practice's behalf is generally a business associate. HHS explains that a covered entity may disclose such information to a business associate only if it obtains "satisfactory assurances, in the form of a contract or other written arrangement," a business associate agreement. A free chatbot you signed up for yourself has no such agreement with your employer.
Massachusetts personal information. Under 201 CMR 17.00, "personal information" means a resident's name, or first initial and last name, combined with a Social Security number, a driver's license or state ID number, or a "financial account number, or credit or debit card number." Businesses that own or license that information about Massachusetts residents must protect it.
Keeping only what you need. The Federal Trade Commission's guide for business puts it plainly: "If you don't have a legitimate business need for sensitive personally identifying information, don't keep it." Pasting data into a third-party tool creates a copy your employer does not control.
Why "I took the names out" is harder than it sounds
Removing a name rarely makes a record anonymous. HHS's de-identification guidance lists 18 types of identifiers that must be removed under its "Safe Harbor" method. Names are only the first. The list also includes:
- All geographic subdivisions smaller than a state, with a narrow exception for ZIP code digits
- All elements of dates except the year, such as a date of service or birth date
- Phone numbers, email addresses, account numbers and medical record numbers
- "Any other unique identifying number, characteristic, or code"
Even then, the information counts as de-identified only if the organization has no "actual knowledge" that what remains could identify someone. A billing clerk deleting a patient's name from a claim note has not met that standard. That decision belongs to your employer's privacy officer, not to you.
If your employer has no policy
- Ask in writing. Email your supervisor: "Which AI tools, if any, am I allowed to use for work, and what data may I put into them?"
- Until you get an answer, use only the "fine" column. Invented examples and public text carry no privacy risk.
- Keep the answer. A written reply protects you if questions come up later.
What if my boss told me to paste it? Ask whether the tool is approved for that data and who approved it. If the instruction stands and you are uneasy, take the question to your employer's privacy or compliance contact.
If you already pasted something
- Stop. Do not paste anything more into that conversation.
- Write down what, when and where: the type of data, roughly how much, the time and which tool.
- Tell your supervisor or privacy officer the same day. Your employer, not you, decides whether it must be reported and to whom. Waiting makes the problem harder to fix.
- Do not try to cover it up by deleting the chat and saying nothing. Deleting a chat on your side may not remove the data from the provider.
When to ask compliance
Ask before you act whenever any of these is true: the data involves health, money or identity documents; you plan to upload a file instead of typing a question; the AI feature is new inside software you already use; or a customer asks you to use AI on their information.
To practice this rule, build a work sample that uses invented data only. The one-page work sample guide shows how, and the ed.cafe 12-point rubric scores "zero private data" as one of its four criteria.
Sources
- Guidance regarding methods for de-identification of PHI, U.S. Department of Health and Human Services, accessed 15 October 2026
- Business associates, U.S. Department of Health and Human Services, accessed 15 October 2026
- 201 CMR 17.00: Standards for the protection of personal information of residents of the Commonwealth, Mass.gov, accessed 15 October 2026
- Protecting personal information: a guide for business, Federal Trade Commission, accessed 15 October 2026
- TEN 07-25: AI Literacy Framework, U.S. Department of Labor, Employment and Training Administration, 13 February 2026.pdf)
- Microsoft: more workers using AI, but they're ashamed to admit it, PCWorld, 8 May 2024
- AI for workers and learners: 2026 survey, Jobs for the Future, accessed 15 October 2026
