The Hidden Dangers of Building Your Own Healthcare App
AI coding tools can build a patient portal, an intake app or a scheduling tool in an afternoon. In 2026, researchers found thousands of these apps leaking sensitive data, including medical records, and one practice's self-built patient app was fully readable within half an hour. What a practice takes on when it builds and runs its own healthcare software, the cases that show what goes wrong, and the skills it really takes.
AI coding tools have made it possible for almost anyone to build software. Describe a patient intake app, a portal or a scheduling tool, and a working version can be on the internet the same afternoon. For a dental or medical practice, that speed is tempting. It is also where the trouble starts, because the hard part of healthcare software was never writing the code. It is keeping patient data safe, private and within the law every day after the app goes live.
This guide is for practice owners, office managers and anyone thinking about building their own patient software, or who already has. It covers the cybersecurity, privacy, HIPAA and compliance risks, the recent cases that show what goes wrong, the skills it takes to build and run healthcare software safely, and how Patientfy AI approaches the same problems. It is general information, not legal advice.
Why this is a 2026 problem
The tools changed faster than the rules. An AI agent will happily build a login screen, a database and an upload page. It will not ask whether the database is open to the public, whether the hosting company signs a Business Associate Agreement, or whether the practice wrote down who is allowed to see what. Those are the questions regulators, insurers and plaintiffs' lawyers ask after a leak.
What can go wrong: real cases
A practice built its own patient app
In March 2026, a security researcher in Switzerland described a medical practice that used an AI coding agent to build its own patient management app, loaded its real patient records into it and put it on the internet. By his account, he could read the full patient record set, voice recordings included, within about 30 minutes. The practice had chosen not to buy regulated patient software. No attacker was confirmed, but the exposure itself was the problem.
Thousands of AI built apps left the door open
In May 2026, the security firm RedAccess reported finding about 380,000 publicly reachable apps built with AI coding tools, and roughly 5,000 of them exposing sensitive data. Reported examples included patient conversations at a children's care facility and doctor and patient summaries from a hospital app. A 2025 scan by another firm found 175 personal data exposures among 5,600 AI built apps, including medical records.
The most common cause is simple. Many AI built apps let the browser talk straight to the database, protected only by access rules the builder has to set. When those rules are missing or wrong, anyone holding the app's public key, which ships inside the page, can read or change every table. In 2025 a researcher found this in more than 170 apps built on one popular AI platform, a flaw now listed in the national vulnerability database as CVE-2025-48757.
- apikey
- eyJhbGciOiJIUzI1NiIs…Shipped inside the page code, so anyone has it
- auth
- noneNo sign-in needed: the access rules were never set
- first_name
- Jordan
- dob
- 1984-03-17
- phone
- (555) 010-0187
- reason
- Root canal follow-upHealth information, readable by anyone
- insurance_id
- XQW482019331
- rows
- 2,481
An AI agent deleted a live database
In July 2025, an AI coding agent on a popular platform deleted a company's production database during a declared code freeze, then gave misleading answers about whether it could be restored. The data was business contacts, not patients. The lesson carries over: an agent with access to live data can destroy it, and a practice needs backups, separate test and live systems, and people who know how to recover.
Software vendors are held to the same rules
On March 5, 2026, HHS announced a settlement with MMG Fusion, a software company serving dental practices, over a breach in which the health information of about 15 million people was posted online. The company agreed to a corrective action plan monitored for three years. Earlier, the 2024 ransomware attack on Change Healthcare affected about 193 million people, the largest healthcare breach on record, and disrupted payments and prescriptions across the country for weeks. Small or large, whoever runs the software carries the duty to protect what is in it.
| Case | What happened | What it shows |
|---|---|---|
| A practice's self-built patient app (2026) | Real patient records readable online | Building it yourself makes you the security team |
| AI built apps (RedAccess, 2026) | About 5,000 apps exposing sensitive data | Default settings are not safe settings |
| CVE-2025-48757 (2025) | 170+ apps with open databases | One missing access rule exposes everything |
| AI agent deletes database (2025) | Live data wiped during a code freeze | Backups and separate systems are not optional |
| MMG Fusion settlement (2026) | 15 million dental patients' data posted online | Software vendors answer to HHS too |
| Change Healthcare (2024) | About 193 million people affected | One weak point can stop a whole practice |
The four risks every healthcare app carries
1. Cybersecurity
Healthcare is one of the most attacked industries. In the first half of 2026, HIPAA Journal counted 397 breaches of 500 or more records, affecting almost 33.8 million people, and nearly all of the largest were hacking or ransomware. A practice app needs strong sign-in with a second factor, encryption, access rules on every table and file, software updates, logs someone actually reads, backups that are tested, and a plan for the day an account is taken over.
2. Privacy
Privacy is about what the app shares, not only what it protects. Analytics tools, chat widgets and error trackers added to an app can send what patients do to outside companies. Washington's My Health My Data Act and similar state laws cover consumer health data even outside HIPAA, and the FTC's Health Breach Notification Rule covers many health apps that HIPAA does not, with penalties per violation.
3. HIPAA
When a covered practice puts patient information into software, HIPAA applies to all of it: the app, the hosting, the database, the email service, the AI model the app calls and every other vendor that touches that information. Each of those vendors needs a Business Associate Agreement, and many consumer and startup plans do not offer one. HIPAA also requires a written risk analysis, a designated privacy and security official, workforce training, access controls and audit logs.
4. Compliance after launch
The work does not end when the app goes live. A breach of patient information must be reported to patients within 60 days of discovery, to HHS, and to the media when 500 or more people in a state are affected. HHS's Risk Analysis Initiative has made the missing risk analysis its main target, and it was the common thread in the settlements HHS announced in 2026, including a $103,000 settlement with a treatment center in February. HHS's proposed Security Rule update would also make encryption and multi-factor authentication required outright. It is not final yet, but it shows where the rules are heading.
- You find and sign a BAA with every vendor
- You set every access rule on every table
- You write and update the risk analysis
- You watch the logs and answer the alerts
- You notify patients and HHS after a breach
- BAAs are part of the service
- Access rules are designed, tested and reviewed
- Security reviews are part of how it runs
- A team watches and answers
- You still own patient notices, with a partner who has a plan
The skills it takes to build and run healthcare software
Writing the first version is the smallest part of the job. Here is what a practice takes on when it builds and operates its own patient software.
Most practices do not have these people on staff, and an AI tool cannot fill those roles. It can write code that looks finished while the access rules underneath are wide open. Before building, ask a simple question: who will be responsible for this app at midnight on a holiday, and do they have these skills?
- List every place patient information will be stored, sent or shown
- Confirm every vendor will sign a Business Associate Agreement
- Write the risk analysis before real patient data goes in
- Turn on access rules for every table and file, and test them while signed out
- Require a second sign-in factor for everyone with access
- Keep test data and live patient data in separate systems
- Set up backups and test a restore
- Name the person responsible for security and for breaches
- Remove analytics, chat and error tools that can see patient data
- Plan who reviews every change before it goes live
How Patientfy AI is aligned
Patientfy AI builds and runs the patient facing parts of a practice's marketing and intake, from websites and ads to Patientfy Consult and the CRM, so these same questions apply to everything it ships. Patientfy AI is built for HIPAA and each state's privacy laws.
No platform removes every risk, and every practice still owns its own HIPAA program. The difference is that a practice working with Patientfy AI is not the one writing the access rules, watching the logs and signing a dozen vendors' agreements alone. For a practice that has already built something itself, the guide to building a practice website with AI covers the website side of the same problem.
Frequently asked questions
Can I build my own HIPAA compliant app with AI?
An AI tool can write the code, but no tool makes an app HIPAA compliant on its own. HIPAA asks who can see patient data, how it is protected, what happens when something goes wrong, and which vendors sign a Business Associate Agreement. Those answers come from how the app is hosted, configured, tested and run every day, and from the people responsible for it.
What happens if my self-built practice app leaks patient data?
A leak of protected health information is usually a reportable breach. The practice has to notify affected patients without unreasonable delay and within 60 days, report it to HHS, and notify the media when 500 or more people in a state are affected. State laws can add their own notices and penalties, and patients can sue. HHS also looks at whether the practice ever did a risk analysis.
Does HIPAA apply to a health app a practice builds?
If the practice is a covered entity and the app creates, receives, stores or sends patient information for it, the app is part of the practice's HIPAA program, and every vendor behind it that touches that information needs a Business Associate Agreement. Health apps outside HIPAA can still fall under the FTC's Health Breach Notification Rule and state consumer health privacy laws.
What is the most common security mistake in AI built apps?
Leaving the database open. Many AI built apps talk to their database straight from the browser, and if the access rules are missing or wrong, anyone with the app's public key can read or change every record. In 2025 a researcher found more than 170 apps built on one popular AI platform with this problem.
Is the HIPAA Security Rule changing?
HHS proposed a major update in January 2025 that would make encryption and multi-factor authentication required and spell out asset inventories and data maps. As of November 2026 it is still a proposal, with final action listed for July 2027. The current Security Rule is fully in force, and HHS is actively settling cases with practices that never did a risk analysis.
How does Patientfy AI handle patient data?
Patientfy AI is built for HIPAA and each state's privacy laws. It works under a Business Associate Agreement with the practices it serves, keeps Patientfy Consult patient details in their own separate database, its own tools do not record visits to practice websites or Patientfy Consult pages, and checks every website it launches each day against what that website's privacy policy promises.
The parts of the platform that do this for your practice, automatically.
Sources
- Axios: AI vibe coding apps leak sensitive data (May 7, 2026)
- eWeek: Thousands of AI vibe coding apps may expose sensitive medical and business data
- Vibe Graveyard: A doctor vibe coded a patient app (March 2026)
- NIST National Vulnerability Database: CVE-2025-48757
- The Register: Replit's AI agent deleted a production database (July 2025)
- HHS: OCR settles HIPAA investigation of MMG Fusion, breach affecting 15 million (March 5, 2026)
- HIPAA Journal: H1 2026 Healthcare Data Breach Report
- TechTarget: Change Healthcare data breach victim count rises to 193 million
- Sidley: Risk analysis in the crosshairs (June 2026)
- Hunton: OCR settles with Top of the World Ranch Treatment Center for $103,000
- Federal Register: FTC Health Breach Notification Rule (May 30, 2024)
- HHS: HIPAA Breach Notification Rule
- HHS: HIPAA Security Rule proposed update (January 2025)
Every day, a patient picks someone.
You decide if it’s you.
Tell us about your practice and we will show you how the platform works. 20 minutes, no pressure, no pitch. Just a real look at what Patientfy can do for you.

