On September 21, an employee at Johns Hopkins Bayview Medical Center in Baltimore uploaded a spreadsheet to a public AI tool the hospital hadn’t approved. The employee was doing an internal review. The spreadsheet held admission and treatment records for about 3,182 patients from May through September 2026, according to a Johns Hopkins release reported by Becker’s Hospital Review.
The hospital found out the same day. It had the employee delete the prompt, the conversation, and the file from the tool’s history, and it asked the company behind the tool to delete the data too. It still notified every affected patient. Letters went out starting October 22 with an offer of 12 months of credit monitoring, and the story ran on CBS Baltimore and in the Baltimore Sun.
Nobody hacked anything. One person had a spreadsheet and a task, and an AI tool was a browser tab away. Every medical and dental practice in Greater Sacramento has staff in that same position: a biller reformatting an aging report, a front desk coordinator cleaning up a recall list, a practice manager summarizing a month of appointments. Each of those takes seconds to paste into a chatbot.
Why This Counts as a Breach
The spreadsheet contained names, admission and treatment dates, medical record and account numbers, departments, medications, and provider names. That’s protected health information under HIPAA, even without Social Security numbers.
Under HIPAA, a practice can share patient information with an outside company that handles it on the practice’s behalf only if that company has signed a business associate agreement (BAA). A public AI tool used through an employee’s own account has no BAA with the practice, so sending patient data into it is generally treated as an impermissible disclosure. Whether it becomes a reportable breach depends on the facts, any applicable exceptions, and a documented risk assessment.
The HIPAA Breach Notification Rule then presumes that disclosure is a breach. The only way out is a documented risk assessment showing a low probability that the information was compromised, weighed across four factors:
How identifiable and how sensitive it is. Names plus medications and dates is a lot.
An AI company with no obligation to the practice under HIPAA.
Once uploaded, the data was processed by the tool. Whether people at the vendor could see it depends on the vendor’s terms.
A deletion request helps, but the practice usually can’t verify that deletion happened.
Johns Hopkins caught the upload the same day and asked for deletion, and it still notified patients. A small practice in a similar position should plan for the same outcome. HIPAA treats size in two separate ways. A breach affecting 500 or more individuals must be reported to HHS within 60 days of discovery; smaller breaches are logged and reported annually. Separately, a breach affecting more than 500 residents of a single state or jurisdiction also requires notice to prominent media serving that area. That’s how a hospital spreadsheet becomes a local news story.
How to Prevent It at a Small Practice
Training alone doesn’t solve this. Staff use AI because it saves them time, and a rule that only says “don’t” pushes the work onto personal phones where nobody can see it. Prevention works best in three parts: give people an approved tool, block the ones that aren’t approved, and put the rule on one page.
1. Give staff an approved AI tool with a BAA
Several major AI products now offer a BAA, but only on specific business plans, and coverage usually stops at certain features. Standard consumer subscriptions generally do not provide BAA coverage. Eligibility depends on the specific product, account configuration, and covered functionality. As of this writing:
Coverage is feature-specific, so check each vendor’s current eligibility list rather than relying on a plan name. A BAA is the starting point. The practice still has to configure the tool, limit who has access, and confirm which features the BAA actually covers. If your practice already runs on Microsoft 365 or Google Workspace, the most practical approved tool is often the one built into the system you already pay for. Our guide to what employees can and can’t share with AI walks through the tool categories in more detail.
The same rule applies to AI scribes and meeting note takers. A tool that listens to a patient visit or joins a telehealth call is handling protected health information and needs a BAA. Check for note-taking bots that staff may have connected to their calendars on their own.
2. Block the AI tools you haven’t approved
Most practices already have, or can easily add, web filtering on office computers. Your IT provider can block known consumer AI sites and browser extensions on practice devices and allow only the approved tool. It won’t stop someone using a personal phone, but it removes the easiest path on the computers that have access to your practice management system. Monitoring can also show which AI tools staff are already using, which is usually more than owners expect. Shadow AI is already inside most organizations, and healthcare is no exception.
3. Put the rule on one page
Staff will remember a short rule better than a long policy. A one-page version should cover the essentials:
- Patient information only goes into the AI tool the practice has approved, using your work account.
- No patient information in personal AI accounts, free AI tools, browser extensions, or apps on personal phones.
- “Patient information” includes names, dates, record numbers, schedules, billing details, and anything else that could identify a patient.
- If you aren’t sure whether something is allowed, ask before you paste.
- If patient information goes into the wrong tool, tell the privacy officer right away. Reporting quickly always gives the practice more options.
Have every staff member read and sign it, and go over it again at your annual HIPAA training. Your HIPAA risk analysis should also list AI tools as a risk area, since the Security Rule expects the analysis to cover how patient information actually moves through the practice.
If It Happens: The First 48 Hours
The clock starts when the practice discovers the disclosure. HIPAA requires notification without unreasonable delay and no later than 60 calendar days after discovery. The decisions that shape the outcome, however, often happen in the first two days.
Document before deleting. Record which tool was used, which account, when, and exactly what was uploaded. Take screenshots of the conversation and keep a copy of the file. You’ll need this for the risk assessment, and it disappears once the history is deleted.
Tell the privacy officer. Every practice has to have one under HIPAA, even if it’s the practice manager wearing another hat. Involve your healthcare attorney early.
Delete and request deletion. Remove the conversation and file from the account, then contact the AI company in writing to request deletion and ask for written confirmation. Keep copies of everything.
Determine whether the tool was covered. If the upload went into an approved tool under a BAA, the analysis may be very different. If it was a personal or free account, assume a breach analysis is needed.
Start the four-factor risk assessment and count affected patients. Write it down as you go. The number of patients determines whether HHS and media notice are required, and the documentation is what you’ll rely on if regulators ask.
Check California deadlines and call your insurer. California adds its own rules, and they don’t apply the same way to every practice. Clinics, health facilities, home health agencies, and hospices licensed by the state have a much shorter window than HIPAA: 15 business days to report to the California Department of Public Health and affected patients under Health and Safety Code 1280.15. Many private medical and dental offices aren’t licensed facilities under that section, so confirm with counsel which state requirements apply to yours. Notify your cyber insurer early, since many policies require prompt notice.
For more on how California’s timelines work, see our overview of California’s breach notification requirements.
Not legal advice. This article provides general information about HIPAA and AI tools for medical and dental practices. Whether a specific incident is a reportable breach, and which state requirements apply, depends on the facts and should be decided with your privacy officer and healthcare counsel. AI vendors change their plans and BAA terms often; confirm current coverage directly with each vendor before relying on it. Last reviewed: October 2026.
Do you know which AI tools your staff are already using?
In 15 minutes, we’ll talk through how AI is likely showing up at your practice, which tools could be approved under a BAA, and what it would take to find and block the rest.
Schedule a 15-Minute CallVision Quest Cyber works with medical and dental practices across Greater Sacramento on AI governance, HIPAA risk analysis, and the security controls that keep patient information where it belongs.
Sources: Becker’s Hospital Review, “Johns Hopkins alerts 3,182 patients after worker uploads data to unapproved AI tool” (Oct. 26, 2026): beckershospitalreview.com • CBS Baltimore (Oct. 26, 2026): cbsnews.com • HHS, HIPAA Breach Notification Rule: hhs.gov • HHS, Business Associates: hhs.gov • Google, HIPAA Included Functionality: workspace.google.com • OpenAI, HIPAA-eligible products and functionality: help.openai.com • Anthropic, BAAs for commercial customers: support.claude.com • CDPH All Facilities Letter 14-26 (H&S Code 1280.15): cdph.ca.gov
