The DOJ Data Security Program treats heart-rate, sleep and blood-oxygen readings as personal health data. It restricts access from China once a data set covers more than 10,000 US persons. If your parent company is in China, its engineers there are generally covered persons. De-identifying or encrypting the data does not change that. What you control is the architecture, meaning where the data lives, who can reach it and what gets logged.
On a patient iPad app for a regulated at-home therapy device, our engineers forked the WatermelonDB sync library to add SQLCipher encryption without losing its sync engine.
The DOJ Data Security Program counts almost every reading a connected-health device takes as personal health data. The definition covers vital signs, bodily functions, symptoms, test results and logs of exercise habits, among other health attributes (28 CFR 202.241).
You do not have to be a hospital for this to apply. The DOJ's own FAQs say the definition applies no matter who collects the data. Their example is "logs of exercise habits, which could be collected by fitness apps" (DOJ Data Security Program FAQs, question 31).
A connected-health app usually carries three more of the rule's data categories.
Precise geolocation is data that places a person or device within 1,000 meters (28 CFR 202.242). Biometric identifiers are traits such as facial images, voice prints, fingerprints and gait that are enrolled in a system to recognize a person (28 CFR 202.204). A heart-rate reading is health data. It would count as a biometric identifier only if a system enrolled it to recognize or verify the person. Covered personal identifiers are listed identifiers linked to each other or to other sensitive data (28 CFR 202.212). Listed identifiers include a device's MAC address, an IP address, an advertising ID and an account username (28 CFR 202.234).
One of the rule's own examples is a device MAC address linked to a home address.
| What the app collects | Example product | Category under the rule | Bulk threshold |
|---|---|---|---|
| Heart rate, sleep stages | Smartwatch, fitness band | Personal health data | More than 10,000 US persons |
| Blood-oxygen readings | Connected pulse oximeter | Personal health data | More than 10,000 US persons |
| Workout history | Connected bike, rower or treadmill | Personal health data (logs of exercise habits) | More than 10,000 US persons |
| GPS route of a run or ride | Watch app with location | Precise geolocation | More than 1,000 US devices |
| Face or voice samples your servers enroll to verify identity | Any app with its own biometric login | Biometric identifiers | More than 1,000 US persons |
| Device MAC address plus account email or IP address | Any paired device | Covered personal identifiers | More than 100,000 US persons |
The thresholds come from 28 CFR 202.205. They count data held at any point in the preceding 12 months, added up across all deals between the same US company and the same foreign counterparty. A set that mixes categories, or links a listed identifier to health data, is "combined data," and the rule measures it against the lowest threshold of the categories inside it.
For a wearable app that records location, the 1,000-device location threshold comes into play long before the 10,000-person health threshold. A watch app that stores run routes next to heart rate is combined data.
We built that patient iPad app, including its Bluetooth device flows, as part of our iOS app development work. For any connected-health app, sorting each field into these categories is the first planning step.
Yes. The rule counts bulk sensitive personal data "regardless of whether the data is anonymized, pseudonymized, de-identified, or encrypted" (28 CFR 202.206). The DOJ's FAQs give the reason. Anonymized data, once aggregated, can still be used to identify people (FAQ question 22).
Encryption still matters. It is one of the four data-level controls CISA lets you combine, but it does not take the data out of scope. At-rest encryption was a hard requirement on the therapy-device app, and the SQLCipher fork of WatermelonDB we built for it is open-sourced.
Once you know your data is in scope, the next question is who counts as a covered person.
Under the DOJ rule, your US company carries the obligations, and your parent company in China, plus the non-US staff it employs there, are generally covered persons.
A company organized solely under US law is a "US person" even when a Chinese parent owns it (28 CFR 202.256). Covered persons include foreign entities organized or principally based in a country of concern, and foreign entities 50% or more owned by them. They also include those entities' foreign employees and contractors, and foreign individuals primarily resident in a country of concern (28 CFR 202.211). China is one of six countries of concern (28 CFR 202.601).
| Who | Status under the rule | Source |
|---|---|---|
| Your US subsidiary, incorporated in the US | US person, carries the obligations | 28 CFR 202.256 |
| Your parent company in China | Covered person | 28 CFR 202.211(a)(1) |
| An engineer employed by the parent and working in China (not a US citizen or green-card holder) | Covered person | 28 CFR 202.211, Example 4 |
| A freelance developer who lives mainly in China (not a US citizen or green-card holder) | Covered person | 28 CFR 202.211(a)(4) |
| A cloud, analytics or support vendor headquartered in China | Covered person | 28 CFR 202.211(a)(1) |
A US citizen or green-card holder working in China is a US person, not a covered person (28 CFR 202.256, Example 2).
Structure matters here. A US branch of a Chinese company is treated as part of its parent, so the DOJ's FAQs say it is not a US person under the rule (FAQ question 51). Whether you run the US business as a branch or a separately incorporated company is a question for your US lawyer. So is the status of any specific person or vendor.
Before any contractor gets access to production data, check where they live and who employs them, along with the checks in our guide to hiring app developers.
Once you know who is covered, the type of arrangement decides whether the rule bans it, restricts it or exempts it.
The DOJ Data Security Program bans two kinds of deals with covered persons outright, restricts three more and exempts some routine work, including intra-company support. Most engineering work a China-based team does for the US company falls in the restricted group. Selling or licensing the data to the parent does not, because the rule treats that as banned data brokerage. The prohibitions and restrictions have applied since April 8, 2025 (DOJ).
The first banned deal is data brokerage with a covered person, meaning the sale or licensing of the data (28 CFR 202.301).
The second is giving a covered person access to bulk human 'omic data (28 CFR 202.303).
Selling bulk data to any other foreign buyer requires a contract that bars resale to a covered person. You also have to report it if you learn the buyer broke that term (28 CFR 202.302). The restricted deals are vendor agreements (cloud services included), employment agreements and non-passive investment agreements that give a covered person access to bulk data. They can go ahead only if the US company meets CISA's security requirements and the rule's due diligence, audit and recordkeeping duties (28 CFR 202.401).
The exemption covers data flows between a US company and its affiliates in China when they are ordinary admin or support work, such as HR, payroll and customer support (28 CFR 202.506).
| Arrangement | How the rule treats it | Source |
|---|---|---|
| The parent's engineers in China log in to production to debug the app | Usually restricted once the data is bulk, and can be effectively prohibited if the work cannot be done without readable, identifiable data | 28 CFR 202.401, 202.258 and FAQ question 69 |
| The US company sells or licenses a bulk data set to its parent in China to help the parent develop its own technology | Prohibited data brokerage | 28 CFR 202.214, Example 6 |
| A vendor headquartered in China processes and stores bulk location data from your app | Restricted vendor agreement | 28 CFR 202.258, Example 1 |
| Part of the app's development is contracted to a covered person whose work involves access to the bulk data | Restricted vendor agreement | 28 CFR 202.258, Example 4 |
| As part of selling ad space, your app passes the ad IDs and IP addresses of more than 100,000 US users' devices in a 12-month period to an ad exchange based in a country of concern | Prohibited data brokerage | 28 CFR 202.214, Example 4 |
| A support agent at an affiliate in China opens individual customer tickets | Exempt as customer support, when it is ordinary support work | 28 CFR 202.506(a)(2)(viii) |
| Bulk data goes to an affiliate's R&D team in China to develop a product | Not exempt | 28 CFR 202.506, Example 4 |
The first row and the support-ticket row are our reading of the closest parts of the rule. The other five rows are the rule's own examples. For your own setup, your US lawyer makes the call.
Two details in the rule's examples catch app teams. First, a company that swaps in its own controls instead of CISA's has not met the condition, and the rule treats that vendor deal as prohibited (28 CFR 202.401, Example 2). Second, the controls do not take a deal out of the rule. Access is judged "without regard for the application or effect of any security requirements" (28 CFR 202.201), so the compliance program, audit and records still apply (FAQ question 68).
DOJ says it expects most restricted deals to go ahead once the controls are in place. The exception is the rare case where a covered person's only service requires readable, identifiable data, which can leave the deal effectively prohibited without a specific DOJ license (FAQ question 69). Design the work so nobody in China needs that.
Violations fall under IEEPA, the statute behind many US sanctions programs, and willful violations carry criminal penalties (28 CFR 202.1301).
A restricted deal can go ahead only if the US company meets CISA's security requirements, and most of that document reads like an engineering spec.
The CISA security requirements ask for two layers of controls. One layer covers every system that touches covered data, and the other covers the data itself. CISA issued them in January 2025 (CISA) under Executive Order 14117, the order the DOJ program implements (DOJ FAQs). The DOJ rule adopts them by reference (28 CFR 202.248). The two tables below summarize the CISA requirements document.
| CISA requirement | What it means in an app stack |
|---|---|
| An asset inventory with each asset's IP address, updated at least monthly | A live list of every server, database, storage bucket and container that holds the data |
| A named person accountable for security, and for governance, risk and compliance | One accountable owner, such as a CISO |
| Known exploited vulnerabilities on internet-facing systems fixed within 45 days | A 45-day patch window your release process has to meet |
| Documented vendor agreements, a network map, approval before new hardware or software goes in, and an allowlist of approved hardware and software | No unreviewed SDK, service or library reaches a covered system |
| Incident response plans for covered systems, reviewed every year | A written runbook for a breach or for access by a covered person |
| Multifactor authentication on all covered systems | MFA on cloud consoles, databases and admin panels |
| Identities and credentials for users, services and hardware, issued and managed centrally | One directory for people, services and devices, with each account limited to the functions it needs |
| Access removed promptly when someone leaves or changes roles | Same-day offboarding, the example CISA gives |
| Central logs of access and security events, kept at least 12 months and readable only by authorized users | One log store, plus an alert when a log source stops sending |
| Deny-by-default connections | Nothing reaches a covered system unless it is explicitly allowed |
| An internal data risk assessment, reviewed every years | A written case for why your controls keep covered persons out |
At the data level, CISA lets you combine four tools. Together they must keep covered persons away from data that is "linkable, identifiable, unencrypted, or decryptable" with commonly available technology.
| Control | What CISA requires |
|---|---|
| Minimization and masking | A written retention and deletion policy, plus processing such as aggregation or pseudonymization before a covered person sees the data. Aggregates must be built from at least as many records as the bulk threshold. |
| Encryption | Covered data encrypted in transit and in storage, with TLS 1.2 or higher. Keys may not sit with the data, may not be stored in a country of concern, and covered persons may not have access to them. |
| Privacy-enhancing technologies | Methods such as homomorphic encryption or differential privacy, applied so covered persons cannot reconstruct the data |
| Identity and access management | Access rules set to deny covered persons access to covered data in every covered system |
A US company in a restricted transaction also needs a data compliance program, required since October 6, 2025 (28 CFR 202.1001). The program has to verify and log data flows in an auditable way, including data types and volumes, who the parties are, and how the data is used and transferred. It needs written policies that a responsible officer certifies every year. An independent auditor, who cannot be a covered person, must audit it every calendar year (28 CFR 202.1002).
The records have to stay available for at least 10 years (28 CFR 202.1101).
Our PM on the therapy-device build advises any PM starting a healthcare project to over-document every decision. "In medical, 'we discussed it on a call' doesn't survive a regulatory or liability question six months later."
Read that list as an engineer and most of it is logging, access rules and documentation the system has to produce. That is why the architecture decides how heavy this compliance work gets.
Build it so the honest answer to "who in China can reach identifiable US user data?" is "nobody," and so the logs can prove it. These are the six decisions to settle first when you adapt a connected-health app for the US market.
Host the production database, backups, analytics warehouse and log store in US cloud regions, in an account the US company owns. Keep encryption keys in a separate key management service, also in the US, and give nobody in China permission to use them. That setup follows CISA's key rules, which keep keys apart from the data, out of every country of concern and out of reach of covered persons. If the app runs on servers in China today, moving it is a project of its own.
Our post on scaling infrastructure as your user base grows covers the constraints to lock before you pick the new architecture.
A US region does not settle the question alone. The rule defines access as logical or physical, through any system or network (28 CFR 202.201). A US database with an admin login used from China is still accessible.
The US company should own the cloud organization, the single sign-on directory and the admin accounts. Accounts for staff in China sit in separate groups, and deny policies block those groups from every resource that holds covered data. Every account uses MFA. People who leave lose access the same day. Every access to a covered system writes to one central log, kept at least 12 months.
AppMakers USA's custom software development team builds these backend pieces, including US-hosted APIs, admin tools with role-based access, and the logging pipeline behind them.
A data map is a table that lists every field your app and device collect. For each field, record its category under the rule, where it is stored, which services and SDKs receive it, and which roles can read it. The compliance program's data-flow logs and the yearly audit both start from it (28 CFR 202.1001).
A professional code audit is a good way to check the map against what the code really sends.
A connected device has two pipelines, and they can be split. Firmware builds, Bluetooth protocol specs and test fixtures can come from engineers in China without touching user data. Readings, device IDs and account data flow from the device to the phone app and on to the US backend only.
AppMakers USA's mobile app development team builds the phone side of that pipeline, including Bluetooth Low Energy pairing.
Crash reports and device diagnostics are easy copies to miss. Strip health values, MAC addresses and account IDs from them before any copy goes to a team in China. A MAC address linked to an account is a covered personal identifier (28 CFR 202.212), and linked to health readings it joins a combined data set (28 CFR 202.205).
A staging database seeded from a production copy is still covered data. Engineers in China who use it are reaching that data. Seed staging with synthetic users instead, and test device flows against a mock. We have done this before for a different reason. On the therapy-device project, the hardware team had not written a line of firmware when app development started, and that gap meant a 4 to 6 week delay.
We built a mocked service that mimicked the device's Bluetooth communication, and the app team kept shipping. The same kind of mock lets a firmware team test against the app with no real user in it.
When the product team in China needs numbers, send aggregates. CISA requires them to be built from at least as many records as the bulk threshold, and the system that builds them counts as a covered system (CISA requirements, section II.A).
We recommend keeping identifiable production data away from the team in China entirely, rather than giving the parent's engineers a masked read-only copy. Masking has to be re-checked every time someone adds a field, and CISA says a mix of controls that fails to stop access is treated as invalid from then on. Separation keeps the list of covered systems short.
Analytics, crash reporting, push, attribution and ad SDKs all send data somewhere.
For each one, record who runs it, where it is headquartered and where the data lands. Once the data is bulk, a vendor headquartered in China that processes or stores it turns the integration into a restricted vendor agreement. Ad SDKs face a stricter rule. An SDK you knowingly add that sends bulk data for targeted advertising to a covered person, or to a social media app one owns, is banned data brokerage (28 CFR 202.214, Examples 7 and 8). The ad-exchange example above is where many apps meet the rule first.
For background, see our posts on app analytics tools and on how APIs and SDKs differ.
Not every question on this list is an engineering question, and the split matters for who you hire.
Your US lawyer decides how the rule applies to your company, and your build team makes that answer true in the system.
| Question | Who answers it |
|---|---|
| Is each arrangement with the parent company prohibited, restricted or exempt? | Your US lawyer |
| Should the US business be a subsidiary or a branch? | Your US lawyer |
| Which fields, services and SDKs carry covered data? | Your build team (the data map) |
| Where do the data, backups, logs and keys live? | Your build team |
| Who can reach covered systems, and how do we prove it? | Your build team (identity and logs) |
| Is the compliance program ready for its yearly audit? | Both |
The engineering half is easier to design in than to add to a live app. Our engineer's advice from the therapy-device build was to "validate your encryption story before you commit to a sync library." Hosting, identity and logging follow the same rule. Settle them before the first US user's data lands, because every later change means moving or re-permissioning live data.
Not publicly, as of October 2026. The DOJ's program page lists no enforcement action, no covered-person designations and no advisory opinions, and its FAQs were last updated September 24, 2025. The obligations apply whether or not anyone has been charged.
No. For this rule, "China" includes the Hong Kong and Macau Special Administrative Regions (28 CFR 202.208), so the rule treats Hong Kong parents, vendors and residents the same way as mainland ones. Ask your US lawyer how other federal laws treat a Hong Kong company.
Generally no. The rule's own example says Chinese citizens located in the United States are treated as US persons, not covered persons, unless the Attorney General designates them individually (28 CFR 202.211, Example 2). For a Chinese citizen, covered status follows where they live and who employs them, not their passport.
No. The DOJ's FAQs say the rule's definition of personal health data depends on the data itself, while HIPAA's depends on the type of entity that holds it (FAQ question 31). Consumer health apps can also fall under the FTC's Health Breach Notification Rule, which the FTC applies to many health apps, fitness apps and connected devices outside HIPAA.
It can. The rule requires an annual report, due by March 1, from a US company that is 25% or more owned by a country of concern or a covered person. The duty applies only if that company is in a restricted transaction involving cloud-computing services (28 CFR 202.1103), so ask your US lawyer whether your cloud setup triggers it.
It can, because CFIUS counts identifiable health or geolocation data on more than one million people as sensitive personal data (31 CFR 800.241). The count covers data maintained or collected in the past 12 months, or planned as part of the core product, and it can bring certain foreign investments in the business into review. The DOJ rule exempts that investment agreement from most of its requirements only if CFIUS imposed terms on data access and expressly designated them a "CFIUS action" (28 CFR 202.508).
Every decision in this post starts from the same document, a map of what your app collects, where it lives and who can reach it.
AppMakers USA is a US app development company with 60+ US-based staff, 400+ apps shipped, and offices in Los Angeles, New York, San Diego and Santa Monica. We adapt apps for American users and can build the US-side architecture to support compliance with rules like this one. Company formation, EINs and bank accounts stay with your lawyers and accountants.
Book a 45-minute call in English with our COO, Aaron, or in Mandarin with a Mandarin-speaking member of our team.