Skip to main content
Compliance

AI Tools in Dental Practices: Do Free AI Tools Create HIPAA Risk?

Your front desk asks ChatGPT to tidy a recall letter. A hygienist dictates notes into a free transcription app. The office manager lets Copilot summarise an inbox full of insurance emails. Nobody thinks of it as sharing patient records, and nobody in the practice ever approved it.

The short answer: yes, free AI tools create HIPAA risk when patient information goes into them. HHS treats a cloud service that handles patient data for a practice as a business associate, which needs a signed business associate agreement (BAA). The free consumer plans we checked come with no BAA, so patient data stays out of them. Work with no patient data in it is a different matter.

This guide is written for the owner or office manager who suspects AI is already in use and wants a plan rather than a ban. It shows which plans of the common tools offer a BAA (from the vendors' own pages, dated), what kind of data can go into which kind of tool, how to tell whether text is truly de-identified, and how to run a 30-day read-only pilot that lets the team use AI without putting patient records at risk. It quotes HHS and the HIPAA regulations rather than interpreting them, and it is not legal advice.

In this guide
  1. How AI Is Being Used Inside Dental Offices Today
  2. Can a dental practice use ChatGPT with patient information?
  3. Why Free AI Tools Create HIPAA Compliance Risk
  4. Which AI tools will sign a business associate agreement?
  5. Why “Nothing Was Hacked” Doesn’t Matter Under HIPAA
  6. Why AI Risk Is Accelerating in 2026 and Beyond
  7. What patient data can go into which AI tool?
  8. What does "de-identified" mean before text goes into an AI tool?
  9. How do you run a read-only AI pilot in a dental practice?
  10. What should your staff AI rule say?
  11. What should you do if patient data already went into a free AI tool?
  12. AI Can Help? When Used the Right Way
  13. What to ask your IT provider about the AI tools your team uses
  14. How Legend Networking Helps Dental Practices Stay Compliant
  15. Why Dental Practices Rely on Legend Networking
  16. How Legend Networking Supports Secure, Compliant Dental Practices
  17. Have questions or need guidance?
  18. Frequently asked questions

How AI Is Being Used Inside Dental Offices Today

AI rarely enters a dental office through a decision. It enters through convenience, one helpful shortcut at a time, and usually through tools the practice already pays for or that staff use at home.

Here is what that looks like in a typical week. A team member pastes a patient's chart note into ChatGPT or Google Gemini to make it read more clearly. Microsoft Copilot is asked to summarise a thread of emails that mention patients by name and procedure. Grammarly checks the spelling of a message to a patient about a treatment plan. A dentist records voice notes on a phone and lets a free transcription app turn them into text for the chart. A browser extension offers to "rewrite" whatever is typed into the practice management system's web portal.

From the user's side, each of these is routine office work. From a privacy side, patient information may already be leaving the practice's controlled systems, going to a company the practice has no agreement with, under terms nobody has read.

The pattern is common because the tools are built into everything. AI assistants now sit inside email, browsers, word processors, phones and operating systems, and several are switched on by default or one click away. The same names come up again and again in dental offices:

Where AI shows upTypical toolWhat staff use it forPatient data likely involved?
Browser or app chatbotChatGPT, Google Gemini, Microsoft Copilot (personal account)rewriting letters, summarising notes, answering questionsoften, if notes or letters are pasted in
Email and documentsCopilot in Microsoft 365, Gemini in Google Workspacesummarising inbox threads, drafting repliesyes, when the mailbox holds patient emails
Writing assistantsGrammarly and similar browser extensionsspelling and tone on patient messagesyes, they read what is typed
Voice and transcriptionfree dictation and transcription apps, ambient scribesturning spoken notes into chart textyes, clinical content by design
Inside the dental softwareAI features in the practice management, charting or imaging software the practice already runsnotes, scheduling insights, radiograph analysisyes, it works on the chart itself
Phone and front deskAI phone receptionists and call summarisersanswering calls, booking, call notesyes, callers give names, dates of birth and reasons for visits

None of these uses is wrong in itself. The question this guide answers is which ones are safe with which data, and who in the practice decides.

Can a dental practice use ChatGPT with patient information?

Only on a plan covered by a signed business associate agreement, set up and checked by the practice. OpenAI's help center says only ChatGPT Enterprise or Edu customers with a sales-managed account are eligible for a BAA for ChatGPT, and that it does not offer one for ChatGPT Business. Free, Plus and Pro accounts are outside it.

That answers the question most practices are really asking, which is whether staff can paste a chart note into the ChatGPT account on their phone. They should not. The account has no BAA, the practice cannot see or control it, and the practice cannot get the data back or confirm it was deleted.

There are three situations where ChatGPT can still be useful:

    • No patient data at all. Drafting a generic post-operative instruction sheet, a job advert, a staff training outline or a website paragraph involves no PHI. A free account is a reasonable tool for this, as long as the output is reviewed before it is used.
    • Properly de-identified text. Text with every HIPAA identifier removed is no longer PHI under the de-identification standard. That is harder to achieve with dental notes than it looks; see the de-identification section below before relying on it.
    • A BAA-covered enterprise plan or the API. OpenAI's help center says the API platform can be used with PHI once a BAA is in place, and that ChatGPT Enterprise or Edu customers with a sales-managed account can explore one through sales. That is a purchasing and configuration project for the practice and its IT provider, not something a staff member sets up alone.

Vendor terms change, so check the vendor's own page on the day you decide, and keep a dated copy with your records.

Why Free AI Tools Create HIPAA Compliance Risk

Free and consumer AI tools are built for individuals, not for healthcare practices. When staff type or paste patient information into them, the practice hands that information to a company it has no business associate agreement with, under consumer terms the practice did not negotiate and cannot enforce.

Most consumer AI plans do not offer a BAA at all. HIPAA expects one whenever an outside company creates, receives, maintains or transmits patient information on the practice's behalf. Without it, the practice has no written assurance about how the data is stored, how long it is kept, whether it is used to improve the vendor's models, who at the vendor can see it, or how it is deleted.

HIPAA also does not ask about intent. The staff member who pasted the note was trying to help a patient. That does not change what happened to the data: it left the practice's control and went to a third party that had not agreed to protect it.

What HHS says about a cloud service that holds patient data

HHS's guidance on HIPAA and cloud computing is direct. "When a covered entity engages the services of a CSP to create, receive, maintain, or transmit ePHI (such as to process and/or store ePHI), on its behalf, the CSP is a business associate under HIPAA." (CSP means cloud service provider; an AI chatbot running in a vendor's data centre is one.) HHS adds that "This is true even if the CSP processes or stores only encrypted ePHI and lacks an encryption key for the data."

HHS also answers the practical question. A covered entity may use a cloud service to store or process ePHI, "provided the covered entity or business associate enters into a … business associate contract or agreement (BAA) with the CSP … and otherwise complies with the HIPAA Rules." HHS's business associate page describes the BAA as the "satisfactory assurances, in the form of a contract or other written arrangement … that the business associate will appropriately safeguard the information".

Two practical points follow from HHS's own words. The BAA comes first, before any patient data goes in. And a BAA is not the whole job: HHS also requires that the practice "otherwise complies with the HIPAA Rules", which includes its own risk analysis of the tool. You can read more about the wider rules in our HIPAA IT guide.

Free, business and enterprise plans are different products

The same brand name can cover very different contracts. A personal ChatGPT, Copilot or Gemini account and the business version your practice could license are separate products with separate terms. The practice's question is never "is this tool allowed?" but "is this plan, under this account, covered by a BAA we have signed, and set up the way we need?"

That is why a rule such as "no ChatGPT" fails in practice. Staff hear it as a ban on one website, then use a different assistant built into their browser. A better rule names the approved accounts and makes everything else off-limits for patient data.

Which AI tools will sign a business associate agreement?

It depends on the plan, not the brand. On 22 October 2026 OpenAI offered BAAs for its API and for ChatGPT Enterprise or Edu sales-managed accounts; Microsoft listed Copilot and Copilot Chat as in scope for its enterprise BAA; Google offered a BAA that Workspace administrators accept; Grammarly offered BAAs only on Enterprise plans. Consumer plans fall outside all four.

The table below records what each vendor's own page said when we read it. It names products only as tools your team may already use; it is not an endorsement, and terms change. Confirm on the vendor's page before you buy or configure anything.

Tool your team may useWhat the vendor's page says about a BAA (read 22 Oct 2026)What that means for patient data
ChatGPT Free, Plus, ProOpenAI: "Only ChatGPT Enterprise or Edu customers that have a sales-managed account are eligible for a BAA for ChatGPT at this time."No BAA on these plans. No patient data.
ChatGPT BusinessOpenAI: "Please note that we don't offer a BAA for ChatGPT Business."No BAA. No patient data, even on a paid business plan.
ChatGPT Enterprise or Edu (sales-managed)Eligible for a BAA through OpenAI sales.Possible, once a BAA is signed and the workspace is configured and reviewed.
OpenAI API (used by some dental software vendors)"To use our API platform with PHI, you'll first need a BAA with OpenAI." Most API services are covered, with listed exceptions.Ask any vendor building on it whether its own BAA with you, and its BAA with OpenAI, are in place.
Microsoft Copilot and Copilot Chat in Microsoft 365 for businessMicrosoft Learn lists "Microsoft Copilot, Microsoft Copilot Chat" among in-scope commercial services; its BAA is "available through the Microsoft Online Services Data Protection Addendum by default to all customers who are covered entities or business associates".Possible inside the practice's business tenant, signed in with work accounts, once configured.
Copilot on a personal Microsoft account or Microsoft 365 Personal or FamilyMicrosoft's HIPAA page describes its enterprise products and services; consumer subscriptions are not on its in-scope list.Treat as no BAA. No patient data.
Gemini in Google WorkspaceGoogle: "Administrators must review and accept a BAA before using PHI in Google services." Gmail's Gemini features "are covered as part of Google Workspace with Gemini in HIPAA Included Functionality."Possible in a Workspace account where the administrator has accepted the BAA and checked the Included Functionality list.
Third-party add-ons inside Google WorkspaceGoogle: "Third-party applications including add-ons are not included in the Included Functionality covered by the BAA."Each add-on needs its own review and agreement.
GrammarlyGrammarly: "we can enter into BAAs only for Grammarly for Business Enterprise plans."No BAA on free or individual plans. Keep it off patient messages.
AI features inside your dental software or imagingVaries by vendor and by feature.Ask the vendor in writing whether the feature is covered by your existing BAA or needs a new one.

Two cautions apply to every row. First, a BAA covers a plan and an account, not a person: a staff member's personal login to the same brand is not covered by the practice's agreement. Second, a signed BAA does not make a tool safe on its own. Microsoft's own FAQ asks whether having a BAA ensures an organisation's compliance and answers "No", explaining that the organisation "is responsible for ensuring that you have an adequate compliance program and internal processes in place". The practice still decides who uses the tool, for what, with which settings.

Why “Nothing Was Hacked” Doesn’t Matter Under HIPAA

One of the most common assumptions in dental offices is that a HIPAA problem needs a hacker. It does not. Under HIPAA, an improper disclosure can be a breach with no attacker, no ransomware and no stolen laptop involved.

The Breach Notification Rule defines the starting point. In the regulation's words, "an acquisition, access, use, or disclosure of protected health information in a manner not permitted under subpart E is presumed to be a breach unless the covered entity or business associate, as applicable, demonstrates that there is a low probability that the protected health information has been compromised based on a risk assessment of at least the following factors" (45 CFR 164.402). HHS lists the same four factors:

    1. "The nature and extent of the protected health information involved, including the types of identifiers and the likelihood of re-identification"
    1. "The unauthorized person who used the protected health information or to whom the disclosure was made"
    1. "Whether the protected health information was actually acquired or viewed"
    1. "The extent to which the risk to the protected health information has been mitigated"

Read that against the everyday AI case. A chart note pasted into a free chatbot is information disclosed to a company with no BAA. Nobody broke in, but the presumption still applies, and the practice is the one that has to document why the probability of compromise is low, or treat it as a breach. HHS also says the practice should keep documentation either that the required notifications were made or that notification was not required.

The cost of this kind of event is rarely a headline. It is the time spent working out what was shared, the documentation, possible notifications, and the conversation with a patient who trusted the practice with their records. That is why this guide puts prevention, a written rule and a controlled pilot ahead of everything else.

Why AI Risk Is Accelerating in 2026 and Beyond

AI adoption is moving faster than practice policies. In 2026, AI capability is built into browsers, email, office suites, phones and operating systems, and new features arrive through ordinary updates. A tool that did not exist on the workstation last month can appear in the sidebar this month without anyone installing it.

Four changes make the risk grow rather than settle:

    • Built-in assistants. Copilot, Gemini and similar assistants sit inside software the practice already uses. Whether they are covered depends on how the practice's accounts are licensed and configured, which is rarely obvious to the person clicking the button.
    • Browser extensions. Writing and summarising extensions can read whatever is on the screen or typed into a web form, including patient portals and web-based practice software.
    • AI inside clinical and front-desk software. Ambient scribes, AI radiograph analysis and AI phone receptionists work on patient data by design. They can be the right choice, but only through a vendor with a BAA and a feature the practice has reviewed.
    • Habits from home. Many team members use AI tools every day in their personal lives and reasonably assume they are fine at work. Many dental teams have not been told which AI tools are approved and which are off-limits, so staff fill the gap with their own judgment.

The result is shadow AI: AI tools used for practice work without the owner, office manager or IT provider knowing about them. Shadow AI is a growing privacy risk in dental offices because it spreads quietly and leaves no trail the practice can review. The answer is not to forbid AI, which pushes it further out of sight, but to approve a small set of tools, say plainly what data goes where, and give the team a sanctioned way to try something new.

For the wider security side of AI, including how attackers use it, see AI vendor risk assessment questions and the broader guide to cybersecurity for dental practices.

Futuristic dental technology with AI and IoT integration
Legend Networking | AI Tools in Dental Practices: Do Free AI Tools Create HIPAA Risk?

What patient data can go into which AI tool?

Sort every task by the data it involves, then match it to a tool with the right agreement. Public or practice-only information can go into most approved tools. De-identified text needs a check against the HIPAA identifier list first. Patient information goes only into a tool covered by a signed BAA, approved in writing, and limited to the minimum necessary.

Three data classes

ClassWhat it isDental examplesWhere it can go
A: No patient dataInformation about the practice, the profession or general topicsjob adverts, generic post-op instructions, staff training outlines, website copy, a policy draftany tool on the practice's approved list, with human review of the output
B: De-identifiedText from which every HIPAA identifier has been removed and nothing left could identify the patienta clinical scenario rewritten with no names, dates, places, numbers or photosapproved tools, only after a second person checks the identifier list
C: Patient information (PHI)Anything that identifies a patient together with their health, care or paymentchart notes, appointment details, insurance emails, radiographs, intraoral photos, call recordingsonly a tool covered by a signed BAA, approved for that task, using the minimum necessary

The "minimum necessary" idea comes from the Privacy Rule, which says a practice "must make reasonable efforts to limit protected health information to the minimum necessary to accomplish the intended purpose of the use, disclosure, or request" (45 CFR 164.502(b)). The regulation lists exceptions, including disclosures to a health care provider for treatment. For AI work, the practical habit is simple: give the tool the least patient detail the task needs, even inside a covered tool.

The data-flow and permission matrix

This is the table to print and put next to the office policy binder. It lists the kinds of AI tool a dental office meets, what data each may receive, whether a BAA is usually available, who approves it, and the controls to put in place before use. "Who approves" assumes a small practice; in a larger group the privacy officer and IT lead may be separate people.

Tool typeData class allowedBAA usually available?Who approvesControls before useRisk if misused
Public chatbot on a personal account (ChatGPT, Gemini, Copilot)A onlyNoOffice manager, as part of the approved listwork email not used to sign up; staff told the never-paste listHigh: PHI leaves the practice with no agreement
Business plan of a chatbot without a BAA (for example ChatGPT Business)A onlyNo, per the vendorOwnerpractice-owned accounts; same never-paste listHigh, and easy to mistake for a covered plan
Business or enterprise plan with a signed BAA (ChatGPT Enterprise or Edu, Copilot in Microsoft 365 for business, Gemini in Google Workspace)A, B, and C when approved for a taskYes, by planOwner and privacy officer, with the IT providerBAA signed and filed; work accounts only; MFA; settings reviewed; activity logging onMedium: depends on set-up and staff habits
AI feature inside the practice management or imaging softwareC for the feature's purposeAsk the vendor; it may sit under an existing BAAOwner, privacy officer and the dentist responsible for the clinical workflowwritten confirmation the feature is covered; user permissions checked; output reviewed by a clinicianMedium: clinical accuracy as well as privacy
AI radiograph or imaging analysisC (images)Usually offered by clinical vendors; confirm in writingOwner and treating dentistBAA; confirm how images leave the practice and where they are stored; clinician signs off every findingMedium to high: clinical reliance on the output
Ambient scribe or transcription appC (clinical speech)Clinical scribes often offer one; free dictation apps usually do notOwner and privacy officerBAA; consent approach agreed; audio retention and deletion terms read; notes reviewed before they enter the chartHigh on free apps; medium on a covered scribe
AI phone receptionist or call summariserC (caller details, reasons for visit)Ask the vendor; confirm subcontractorsOwner and privacy officerBAA; recording notice and state recording rules checked with counsel; retention terms; who can listen backHigh: every call carries PHI
Writing or grammar browser extensionA only, unless an enterprise plan with a BAARarely, and only on enterprise plansOffice managerextension allowed only on the approved list; removed from workstations used for patient work if not coveredHigh: it reads whatever is typed
Marketing and image generatorsA only (no patient photos or names)NoOffice managerno patient images or testimonials without written patient authorisationMedium: privacy and consent

If a tool does not appear on the practice's approved list, it is class A only until someone has approved it for more.

What does "de-identified" mean before text goes into an AI tool?

De-identified has a specific meaning under HIPAA. Health information is not individually identifiable when it "does not identify an individual and with respect to which there is no reasonable basis to believe that the information can be used to identify an individual" (45 CFR 164.514(a)). HHS recognises two ways to get there: an expert's determination, or the Safe Harbor method.

For a dental office, Safe Harbor is the practical route. It requires removing a list of identifiers "of the individual or of relatives, employers, or household members of the individual", and the practice must not have "actual knowledge" that what remains could identify the patient. The identifiers include:

    • names
    • geographic units smaller than a state, including street address, city, county and ZIP code (with a narrow exception for the first three ZIP digits)
    • all elements of dates except the year that relate to the individual, such as birth date, admission and discharge dates and date of death, plus all ages over 89
    • telephone numbers, fax numbers and email addresses
    • Social Security numbers, medical record numbers, health plan beneficiary numbers and account numbers
    • certificate or licence numbers, vehicle identifiers, and device identifiers and serial numbers
    • web URLs and IP addresses
    • biometric identifiers, including finger and voice prints
    • full-face photographs and any comparable images
    • any other unique identifying number, characteristic or code

Two points from HHS matter for AI use. First, "The de-identification standard makes no distinction between data entered into standardized fields and information entered as free text": a name inside a chart note counts the same as a name in the patient field. Second, the "actual knowledge" condition means a rare procedure, a small town and a distinctive detail can still identify someone even after the listed identifiers are gone.

Dental text is full of identifiers that are easy to miss. Appointment notes carry dates. Insurance emails carry member numbers. Image files are often named with the patient's name or chart number, and the image itself may carry details in its file information. A voice recording is a voice print. That is why this guide treats de-identification as a checked step with a second pair of eyes, not something a staff member does in their head before pasting.

A simple test before any class B use: could someone in your town who knows the patient recognise them from this text? If the answer is "possibly", it is not de-identified for this purpose. Keep it in class C.

How do you run a read-only AI pilot in a dental practice?

Pick one task, one tool and a few users, and run it for 30 days with patient data kept out unless a signed BAA is in place. "Read-only" means the AI drafts or summarises, and a person reviews and enters anything that reaches the chart, the schedule or a patient. At day 30 the owner decides: adopt, adjust or stop.

A pilot answers the question an owner actually has, which is "would this help us, and can we do it safely?", without committing the whole practice. It also gives the team a sanctioned way to try AI, which reduces the pull towards shadow AI.

Before day 1

  • Pick one task. Choose a single, repetitive job with a clear measure, such as drafting recall and reactivation letter templates, summarising the practice's own policies for new hires, or drafting replies to common non-clinical questions.
  • Set the data class. Write down whether the task is class A (no patient data), B (de-identified, with a second-person check) or C (patient information). Start with class A or B unless a BAA-covered tool is already licensed.
  • Confirm the BAA before any patient data. For a class C pilot, file a copy of the signed BAA and a dated copy of the vendor's page showing the plan is covered.
  • Use practice-controlled accounts. Only accounts the practice owns and an administrator manages, signed in with work credentials and multi-factor authentication. No personal accounts, no shared logins.
  • Review the settings. Record the retention, history and data-use settings the vendor offers for the plan, and set them to the most restrictive option that still works for the task.
  • Turn on logging. Make sure the plan's activity or audit logging is on, and that someone can read it. The Security Rule's audit controls standard asks for "mechanisms that record and examine activity in information systems that contain or use electronic protected health information" (45 CFR 164.312(b)).
  • Name the people. One owner of the pilot, two or three users, and one reviewer for outputs. Write their names on the record sheet.
  • Write the stop rule. Any patient data entered outside the agreed class stops the pilot for that user the same day and is reported to the privacy officer.

Days 1 to 30

  • Users work only on the named task, in the named tool, with the named data class.
  • Every output is reviewed by a person before it is used. Nothing the AI drafts is sent to a patient, entered into the chart or scheduled without a human check.
  • The AI gets no write access to the practice management system, the imaging system, email sending, or the phone system during the pilot.
  • Users log, in one line each, the time saved, errors spotted and anything that felt unsafe.
  • The pilot owner reads the tool's activity log once a week and notes anything unexpected.

The day-30 review

At day 30, the owner, the pilot users and the IT provider sit down for 30 minutes and answer five questions in writing:

    1. Did the tool save time on the named task, and roughly how much per week?
    1. How many outputs needed correction, and were any errors clinically or financially significant?
    1. Did any data go into the tool outside the agreed class? If yes, what was done about it?
    1. Is the plan's BAA, account set-up and logging in a state we would accept for wider use?
    1. Decision: adopt for this task, adjust and run another 30 days, or stop.

Exit criteria

Stop the pilot, and do not extend it, if any of these is true: patient data went into a tool without a BAA; the vendor changed its terms or plan coverage during the pilot; activity logs could not be read; or outputs needed so much correction that no time was saved. Each of those is a clear, documented reason, and that record is useful in itself.

The pilot record sheet

FieldEntry
Task
Tool, plan and account owner
Data class (A, B or C)
BAA on file (date, plan covered)
Settings recorded (retention, history, data use)
Logging on and who reads it
Pilot owner, users, output reviewer
Start date and day-30 review date
Weekly log check (initials, date)
Time saved per week (estimate)
Corrections needed (count, worst example)
Out-of-class data events (count, action taken)
Decision at day 30 and who made it

The HIPAA Security Rule's risk analysis requirement asks a practice to "Conduct an accurate and thorough assessment of the potential risks and vulnerabilities to the confidentiality, integrity, and availability of electronic protected health information" it holds (45 CFR 164.308(a)(1)(ii)(A)). A completed record sheet does not replace that analysis, but it is the kind of dated evidence that feeds it. For a broader structure to manage AI risk over time, NIST's AI Risk Management Framework, "intended for voluntary use", organises the work into four functions: Govern, Map, Measure and Manage.

What should your staff AI rule say?

Keep it to one page that every team member signs. It should name the approved AI tools and accounts, list the data that never goes into any other tool, say who approves new tools, explain how to request one, and tell staff exactly what to do if patient data goes somewhere it should not.

Dental teams do not need a long policy to get this right; they need clarity. A one-page rule works better than a binder because people read it. It can sit alongside the practice's existing privacy and security policies.

1. Approved tools. List each approved tool, the plan, and the data class allowed. For example: "Copilot in our Microsoft 365 business account, signed in with your work account: class A and B. Class C only for the approved task in the current pilot."

2. The never-paste list. Unless the tool is on the approved list for class C, never enter: patient names, dates of birth or appointment dates; chart notes or treatment plans; insurance details, member numbers or claim information; radiographs, intraoral photos or any patient image; phone numbers, email addresses or home addresses; call recordings or voicemails.

3. Accounts. Use only practice-owned accounts for practice work. Do not sign up for AI tools with a practice email address without approval. Do not use a personal AI account for anything involving patients.

4. Extensions and add-ons. Do not install browser extensions or add-ons that read or rewrite text on workstations used for patient work, unless they are on the approved list.

5. Review. A person checks every AI output before it reaches a patient, the chart, the schedule or a payer.

6. New tools. Ask the office manager. New tools start as class A only until approved for more.

7. If something goes wrong. If patient data goes into a tool that is not approved for it, stop, do not delete anything yet, and tell the privacy officer the same day. You will not be in trouble for reporting it; the practice needs to know quickly.

Review the rule whenever a tool or plan changes, and at least once a year alongside the practice's security risk analysis.

What should you do if patient data already went into a free AI tool?

Treat it as a possible impermissible disclosure and document it the same day. Record what was shared, in which tool and account, when and by whom, then bring in the privacy officer. HIPAA presumes such a disclosure is a breach unless a documented risk assessment shows a low probability of compromise, so the practice's records matter.

This section describes what HHS and the regulation say and the practical steps around them. It is not legal advice; for a specific event, involve the practice's privacy officer and counsel.

    1. Stop further use. The staff member stops using the tool for practice work. Do not paste anything else to "check" what it remembers.
    1. Write down the facts while they are fresh. Which tool, which plan, whose account; the date and time; exactly what was entered (copy it from the chart, not from the AI tool); how many patients are involved; and whether anyone else saw the output.
    1. Preserve before you delete. Take a screenshot or export of the conversation for the record if the tool allows, then use the tool's delete or history controls. Deleting a conversation does not undo the disclosure, but it can be part of mitigation, one of the four factors.
    1. Tell the privacy officer. The privacy officer leads the assessment against the four factors in 45 CFR 164.402: the nature and extent of the information and identifiers, who received it, whether it was actually acquired or viewed, and how far the risk has been mitigated.
    1. Read the vendor's terms for that plan. Note what the vendor says about retention, use for model training and deletion, with the date you read it. This feeds factor two and factor four.
    1. Decide and document. The practice either documents why the probability of compromise is low or proceeds with breach notification. HHS says to keep documentation either way.
    1. Close the gap. Add the tool to the never-paste list, check whether other staff use it, and consider whether a covered alternative would serve the task.

Handled calmly and on the same day, a single paste leaves a clear record of what happened and what the practice did about it. Left unrecorded, it becomes a question the practice cannot answer when someone later asks.

AI Can Help? When Used the Right Way

AI tools such as Microsoft Copilot, ChatGPT, Google Gemini and Grammarly are not going away, and many of them can genuinely save a dental team time. The danger is not AI itself. It is unprotected use: patient data in the wrong account, without an agreement, without review.

Used the right way, AI can take real work off the team's plate:

    • Front desk: drafting recall, reactivation and no-show letter templates (class A), and, in a BAA-covered tool, summarising long insurance email threads (class C, approved task).
    • Office manager: turning policies into training outlines and quizzes, drafting job adverts, summarising vendor contracts the practice holds (class A).
    • Hygiene and clinical team: a covered ambient scribe or AI charting feature inside the practice's dental software, with every note reviewed before it is signed (class C, approved task).
    • Dentist and owner: drafting patient-education handouts in plain language for review (class A), and reviewing AI radiograph findings from a covered clinical vendor as a second look, never as the final word (class C).
    • Compliance paperwork: drafting a staff AI rule, a training log template or a pilot record sheet like the one above (class A). AI can help produce documentation; the practice still decides what it says and keeps the evidence.

If you do not yet know how AI is used in your practice, find out first. A quiet look at which tools, extensions and accounts the team already uses is the best starting point for a plan.

What to ask your IT provider about the AI tools your team uses

Your IT provider should be able to tell you which AI features are switched on in your accounts, which plans are covered by a BAA, and how logging is set up. If it cannot answer these questions in writing, the practice cannot show what its AI tools are doing with patient data.

Ask these questions and keep the answers with your security records:

    • Which AI assistants are active in our Microsoft 365 or Google Workspace accounts today, and under which licence?
    • Is our Microsoft or Google BAA in place, who accepted it, and where is the copy?
    • Which browser extensions are installed on workstations that open the practice management system, the imaging software or patient email?
    • Can we limit or remove AI extensions and add-ons on those workstations, and how will we know if a new one appears?
    • Is activity logging on for the AI tools we approve, how long are logs kept, and who reviews them?
    • Are staff signing in to AI tools with practice accounts and multi-factor authentication, or with personal accounts?
    • Which AI features have our dental software and imaging vendors enabled, and are they covered by our existing BAAs with those vendors?
    • If a staff member pastes patient data into an unapproved tool, what can you see, and how quickly can you help us document it?
    • Will you take part in our 30-day pilot review and sign off on the account set-up?

For the deeper questions to put to an AI vendor itself, such as how it trains models, where data is stored and which subcontractors it uses, use the separate AI vendor risk assessment questions. This guide stays with how your team uses the tools.

How Legend Networking Helps Dental Practices Stay Compliant

Staying on the right side of HIPAA with AI is mostly about visibility and set-up: knowing which tools and accounts are in use, making sure the covered ones are configured properly, and making sure the rest never see patient data. That is IT work as much as policy work, and it is where a dental IT provider earns its place.

Legend Networking is a dental IT provider. What we can state about how we work is short and specific: 24/7 monitoring; In-house engineers; no outsourced first-level triage; Onsite dispatch by appointment across served dental markets; remote-first nationwide. Our help desk line is the same for every question: Helpdesk staffed 24/7 at 1-800-794-1588.

A practice planning its AI use can start with the questions in the block above and bring the list of AI tools and accounts it finds to a conversation with us. HIPAA itself is the practice's responsibility; our role is the technology underneath it, described on our HIPAA support for dental practices page.

Why Dental Practices Rely on Legend Networking

Dental practices choose an IT partner on track record and on how the work is done. Here is what we can state:

    • Track record: Served over 3,500 practices since 2004.
    • Reviews: 4.8 rating · 173 Google reviews.

Legend's operations are led by Ian Lynch: 22 years in dental IT; oversees the NOC, onsite dispatch and client technology roadmaps.

Practices work with us to:

    • reduce the technology risks around patient data, including the risks that come with everyday tools such as AI assistants
    • protect patient trust by keeping systems secure, monitored and recoverable
    • plan technology for established practices, new practices and multi-location groups

We support the IT environment of practices running Dentrix, Eaglesoft, Open Dental and Archy, and of imaging systems including DEXIS, Vatech, Carestream, Planmeca/Romexis and Dolphin Imaging.

How Legend Networking Supports Secure, Compliant Dental Practices

Patient care is at the centre of every successful dental practice. Your team should not be distracted by technology problems, security worries or questions about whether a new tool is safe to use.

Legend Networking works on the systems in the background so that technology supports the schedule instead of interrupting it: monitoring, security, backups, workstations and servers, cloud services and the accounts that hold them. AI tools now sit inside those same accounts, which is why the practice's approved list and staff AI rule matter. Our dental cybersecurity services cover the protective side, and the HHS HIPAA home page is the primary reference for the rules themselves.

The aim is to prevent problems before they affect patients, rather than react to them afterwards. For AI, that means a short approved list, accounts the practice controls, logging someone actually reads, and a pilot process that lets the team adopt useful tools without guessing. For the wider rules, our HIPAA IT guide and cybersecurity for dental practices cover the ground around this topic.

Dental practice team meeting with IT consultant
Legend Networking | AI Tools in Dental Practices: Do Free AI Tools Create HIPAA Risk?

Have questions or need guidance?

If your team is already using AI tools, or wants to, the next step is a short, practical review rather than a ban. Bring the list of AI tools and accounts you know about, and ask us for a Free Dental IT Assessment. Tell us what you are running and we will review it with you. Start with our dental cybersecurity services, or call 1-800-794-1588.

Questions

Frequently asked questions

Can a dental practice use ChatGPT with patient information?

Not on a free or ordinary paid account. OpenAI says only ChatGPT Enterprise or Edu customers with a sales-managed account are eligible for a BAA for ChatGPT, and it does not offer one for ChatGPT Business. Without a signed BAA, keep patient information out. ChatGPT can still help with work that contains no patient data, reviewed before use.

Can dental staff use Microsoft Copilot with patient information?

Only inside the practice's Microsoft 365 business account, signed in with work credentials, and after the set-up is reviewed. Microsoft lists Copilot and Copilot Chat as in scope for its enterprise BAA. Copilot on a personal account or a Personal or Family subscription is not on that list, so it should not see patient information.

Can staff use Grammarly on patient emails?

Not on free or individual plans. Grammarly says it can enter into BAAs only for Grammarly for Business Enterprise plans, and asks users not to process protected health information without a current BAA. Because a writing extension reads what is typed, keep it off workstations and accounts used for patient emails unless the practice holds that Enterprise agreement.

What makes AI shadow IT in dental offices?

AI becomes shadow IT when staff use it for practice work without the owner, office manager or IT provider knowing. Assistants built into browsers, phones and email make this easy, often with personal accounts. The practice loses sight of where patient information goes, which is why an approved list and a simple way to request new tools matter.

What are the safe ways for dental practices to use AI?

Start with work that contains no patient data, such as templates, training outlines and policy drafts. Approve a short list of tools and accounts, give staff a one-page rule, and keep patient information for tools covered by a signed BAA. Run a 30-day read-only pilot before wider use, with a person reviewing every output.

Are there free AI tools a dental practice can use with patient data?

We did not find a free consumer AI plan that comes with a BAA among the tools covered here. The vendors' own pages tie their BAAs to business or enterprise plans that an administrator manages. Free tools remain useful for work with no patient data. For patient data, budget for a covered plan and a proper set-up.

Is an AI receptionist safe for a dental office to use with patient calls?

It can be, if the vendor signs a BAA and the practice has reviewed how calls are recorded, transcribed, stored and deleted. Every call carries patient details, so ask which subcontractors handle the audio and transcripts. State call-recording rules differ, so check the notice wording with counsel, and pilot the system before it answers every call.

Does a BAA make an AI tool safe to use?

No. A BAA is the required starting point, not the finish line. Microsoft's own FAQ says having its BAA does not on its own ensure a customer's compliance. The practice still controls who uses the tool, with which accounts and settings, for which tasks, and whether outputs are reviewed and activity is logged.

Can AI help with dental compliance documentation?

Yes, as a drafting aid. AI can turn notes into a first draft of a staff AI rule, a training log, a pilot record sheet or a policy summary, usually as class A work with no patient data. The practice still decides what each document says, checks it against the HIPAA rules and keeps the signed, dated evidence.

Ian Lynch
About the author

Ian Lynch

Owner & Operator, Legend Networking

22 years in dental IT. Oversees the NOC, onsite dispatch and client technology roadmaps for practices nationwide.

  • Operations center

    Dallas, Texas

    Serving dental practices nationwide since 2004

  • NOC hours

    24/7 monitoring

    Helpdesk staffed 24/7 at 1-800-794-1588

  • Engineers

    In-house only

    No outsourced first-level dental IT triage

  • Onsite dispatch

    By appointment

    Technicians deployed across served dental markets

About Legend Networking
Your next step

Want a second opinion on your practice IT?

Tell us what you are running and we will review it with you.

Or call 1-800-794-1588

Free IT assessment

Tell us what you are running

Your information is secure and will never be shared.