A 2016 FTC order defined location to include what an ad network inferred from Wi-Fi data
On June 22, 2016, a complaint filed for the FTC alleged an ad network located consumers through Wi-Fi data after they denied permission. The stipulated order counted such inferences as location, which in our reading made derived data the thing to govern.
Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.
Two-minute summaryThe dek and the first sentence of each section, quoted from the article
On June 22, 2016, a complaint filed for the FTC alleged an ad network located consumers through Wi-Fi data after they denied permission. The stipulated order counted such inferences as location, which in our reading made derived data the thing to govern.
- The week The complaint described the company’s business.
- The inference: when derived data counts as location Location can be read directly from a device’s location interface.
- The signal: reading consent and its limits from the device The second part concerns the person’s choice.
- The host: third-party code inside someone else’s app The third part is the boundary between organizations.
- The proof: showing that controls work over time The fourth part is evidence.
- What the order did not settle The order resolved one company’s case.
- How the issues land in banks, health systems and agencies The four parts apply wherever an organization ships an app with embedded code.
- Objections to this reading Four objections deserve an answer.
- Checks a technology or privacy leader can run now These checks follow from the case’s structure.
- With hindsight The children’s privacy rule the case relied on was amended again.
The week
On June 22, 2016, the United States filed a complaint in the Northern District of California against InMobi Pte Ltd., a Singapore-based company that operates a mobile advertising network. [1] The government acted upon notification and authorization to the Attorney General by the Federal Trade Commission. [1] A stipulated order for permanent injunction and civil penalty judgment was filed the same day for the court to enter. [2] The copy we read is that filed version, before the judge’s signature, so we describe its terms and do not state the date the court entered it. [2]
The complaint described the company’s business. App developers integrate its software development kit, and advertisers can then reach consumers across all the apps that carry it. [1] The complaint said the company reported in February 2015 that its network had reached over one billion unique mobile devices. [1]
The complaint described three geo-targeting products. One targeted consumers by current location. One targeted them by conditions, such as visiting a place at a particular time. One targeted them by location history for up to the last two months. [1]
The complaint then set out how Android and iOS controlled access to location. Both operating systems require apps to obtain the consumer’s consent through permissions before using the location programming interfaces. [1] Both let consumers restrict location access later through system settings. [1]
The core allegation concerned Wi-Fi data. The complaint alleged that, until December 2015, the company tracked consumers’ locations even when they had restricted an app’s access to location. [1] It alleged the company did so by collecting information about Wi-Fi networks the device connected to or that were in range. [1]
According to the complaint, on Android the company collected network names, unique network identifiers called BSSIDs and signal strength when apps included certain Wi-Fi permissions. [1] On iOS, it alleged, the company used an interface intended for captive network authentication to collect the BSSID of each connected network, with no permission dialog shown to the consumer. [1]
The complaint alleged the company built its own geocoder database. Where location access was available, it collected coordinates alongside Wi-Fi information and matched networks to places. [1] Where location access was not available, it fed Wi-Fi information into that database and inferred the consumer’s latitude and longitude. [1]
The complaint said that in response to the Commission’s investigation, the company released a new SDK version in November 2015 and made server-side changes in December 2015. [1] It also alleged that statements in the company’s integration guides and marketing represented that it tracked location only with access to the location interfaces and opt-in consent. [1]
A second set of allegations concerned children. The complaint alleged that thousands of developers had told the company, through a registration option, that their apps were directed to children. [1] It alleged that until October 2015 the company knowingly collected personal information from those apps, including device identifiers, geolocation and BSSIDs, without parental notice or consent. [1]
The complaint charged three deceptive practices under the FTC Act and violations of the Children’s Online Privacy Protection Rule. [1] Under the stipulated order, the company neither admitted nor denied the allegations except as stated in the order. [2]
The stipulated order provided for a civil penalty judgment of $4,000,000. It required payment of $300,000 within seven days of entry and $650,000 in two equal installments, with the rest suspended on the strength of the company’s sworn financial statements. [2] It also required deletion of children’s information and of location information collected or inferred without consent. [2]
The June 2016 case can be read as a story about one company’s broken promise. Its integration guides said location would flow only with permission, and the complaint alleged that it flowed without permission through Wi-Fi data. [1] On that reading, the lesson ends with the company.
Our argument is that the order’s definition reached further. It defined location information to include information about a consumer’s location inferred from other data collected through a programming interface, including BSSIDs. [2] It then barred the company from collecting or inferring location without first confirming consent and the consumer’s settings. [2]
In our reading, that made the inference itself the regulated act. A consumer’s refusal had to govern what the company could work out about them, as well as what it could read from a sensor. For any organization that ships an app with third-party code in it, that is a data governance problem about derived data.
We take the case in four parts, each a separate place where an organization’s control over location data can fail:
- The inference: what counts as location when it is derived from other signals.
- The signal: how consent and its limits are read from the device before any collection or inference.
- The host: what happens when the code doing the collection sits inside someone else’s app.
- The proof: how an organization shows that its controls work, over years.
The parts are distinct. The first is about the data, the second about the person’s choice, the third about the boundary between organizations, and the fourth about evidence. Together they cover the path from a device signal to a record an examiner or regulator could ask for.
After the four parts we look at what the order did not settle, how the issues differ across banks, health systems and public agencies, the strongest objections, and checks a leader can run. The case concerned allegations resolved by a stipulated order, and we describe it that way throughout. [2]
The inference: when derived data counts as location
Location can be read directly from a device’s location interface. It can also be worked out from other signals: the Wi-Fi networks a device sees, the cell towers it uses, or the IP address it connects from. The complaint focused on the second route. [1]
The order’s definition of location information had two branches. The first covered information about a consumer’s location collected through a programming interface. [2] The second covered information about location inferred from any other data collected through such an interface, including BSSIDs. [2]
The city-level carve-out
The definition had one exception. It excluded IP addresses used to infer location at no greater accuracy than city level. [2] That line shows the drafters had to decide how precise an inference must be before it counts.
In our reading, the carve-out tells us the order cared about precision more than source. A city-level guess from an IP address was outside the definition. A building-level fix from a Wi-Fi identifier was inside it. The same logic could apply to any signal that can be turned into a precise position.
For a data team, that means classification by precision. A field that holds a city is different from a field that holds coordinates, even if both came from the same device. Data catalogs that label fields only by their source may miss the difference.
How the inference was built
The complaint described a two-step process. Where consumers allowed location access, the company collected coordinates alongside Wi-Fi identifiers and built a database matching networks to places. [1] Where consumers did not allow access, it used that database to infer coordinates from Wi-Fi identifiers alone. [1]
That process is a simple model trained on data from consenting users and applied to non-consenting ones. The consent of the first group produced the reference data. The second group’s refusal did not stop the model from locating them. In our reading, that is the pattern the order’s definition was written to catch. [1] [2]
Derived data in other settings
The same pattern appears wherever an organization infers a sensitive attribute from less sensitive data. A model might infer where a customer lives from transaction locations, or infer a health condition from purchases. The 2016 order addressed location only. [2]
We do not claim the order governs those other inferences. We do read it as a clear statement, in one domain, that a regulator can treat an inference as the same kind of data as the attribute it estimates. Data governance programs that classify only collected fields leave derived fields unclassified.
Signals that look harmless
A BSSID is a network hardware identifier. On its own it says nothing about a person. Combined with a reference database, it can place a device within a building. The complaint alleged that consumers had no reason to know Wi-Fi permissions would be used to track location. [1]
For a security or privacy review, the question is therefore what can be computed from a permission or field, whatever it is called. That question is harder to answer from documentation. It may require testing what data actually leaves the device and asking what the recipient does with it.
Deleting what was inferred
The order’s deletion terms followed its definition. The company had to destroy location information collected or inferred before the order without meeting the consent conditions, within ten days of entry, subject to exceptions for government requests and legal requirements. [2]
Deleting inferred data is harder than deleting collected data. Inferences may be stored as attributes on profiles, folded into audience segments, or used to train further models. A deletion process that finds raw coordinates may miss the segments built from them.
In our reading, that is the strongest practical argument for tracking lineage. If an organization knows which records were derived from which inputs, it can delete or correct the derived records when the inputs were collected without a valid basis. Without lineage, it cannot be confident it has finished.
Model outputs as records
The geocoder database the complaint described was itself a derived asset. It matched networks to places, using data from consenting users. [1] Assets like that outlive the data that built them, and they keep producing inferences.
An organization using models that infer sensitive attributes should record what data trained them, under what basis, and what the model is allowed to infer about whom. Those records are the derived-data equivalent of a consent log. They let the organization answer a regulator’s or a customer’s question about why it knows something.
Location hiding in ordinary logs
Derived location can also come from systems built for other purposes. Server logs record IP addresses. Analytics tools may convert those addresses to places. Fraud tools may store device network details. Each can create location data that no product owner asked for.
The order’s carve-out for city-level IP inference shows where one regulator drew a line in 2016. [2] An organization reviewing its own logs can ask a narrower question: does any system turn these fields into something more precise than a city, and who can see the result. The answer may lie in the default settings of analytics and security tools.
The signal: reading consent and its limits from the device
The second part concerns the person’s choice. The order barred the company from collecting or inferring location without first confirming three things. [2]
- The consumer had given affirmative express consent for the collection of location information to the app or service that integrated the company’s SDK. [2]
- The consumer had not expressed, through any operating system, device, browser or app permission or setting, that they did not consent or had revoked consent. [2]
- The consumer had not limited consent, through any such permission or setting, to a level of accuracy less precise than what the company would collect or infer. [2]
Consent to the host, not to the SDK
The first condition ties consent to the app that carries the SDK. The consumer agrees with the app they are using. The SDK’s provider relies on that consent. [2]
That arrangement puts the host app in the middle. If the host’s consent screen does not mention location, or mentions it for a different purpose, the SDK cannot rely on it. In our reading, the host app owner therefore controls the basis on which every embedded component may use location.
Settings as continuing instructions
The second condition treats device settings as instructions that can change. A consumer who turns off location in system settings has revoked consent, and the order required the company to respect that. [2]
That requires checking the setting each time, not once at installation. A design that caches a past permission and keeps inferring location after the consumer turns it off would not meet the condition. The complaint said a consumer may restrict location access when, for example, visiting a sensitive location. [1] A setting that is honored only some of the time does not protect that choice.
Accuracy as part of consent
The third condition is the most specific. It treats a consumer’s choice of approximate rather than precise location as a limit the company must honor, including in what it infers. [2]
In our reading, that means an organization cannot take an approximate location the consumer agreed to share and refine it with other signals into a precise one. The refinement would exceed the consent. For engineering teams, that is a rule about combining data sources as well as reading them.
What the order required before acting
The order’s wording is that the company must first confirm these conditions. [2] That places the check before collection or inference, not after. A system that collects everything and filters later would struggle to show it confirmed consent first.
For organizations designing their own apps, the practical form is a gate in the code path: no location-related call or inference without a current consent check. Logs of those checks are the evidence that the gate exists.
Two platforms, two consent models
The complaint described different consent designs. On Android 5.1 and earlier, location permissions were presented when an app was installed, and a system setting could restrict location for all apps. [1] On iOS, a dialog asked the consumer the first time an app sought location, and settings allowed restriction globally or app by app. [1]
An organization shipping on both platforms must read consent from two systems with different timing and granularity. A single consent model in the organization’s own code, mapped to each platform’s signals, makes the gate in the previous section easier to build and test. That is a design suggestion, not a requirement from the order.
Consent and identifiers
The complaint alleged that the information collected from children’s apps included unique device identifiers as well as location data. [1] The 2013 COPPA amendments had brought persistent identifiers that can recognize a user over time and across services into the definition of personal information. [3]
Identifiers and location together are more revealing than either alone. An identifier lets a series of locations be tied to one device. For consent design, that means the choice about location and the choice about tracking identifiers interact, and an organization should treat them together.
The host: third-party code inside someone else’s app
The third part is the boundary between organizations. The company’s code ran inside apps built by others. The complaint alleged that, as a result, app developers could not give consumers accurate information about their apps’ privacy practices. [1]
The complaint said that numerous developers had told consumers in their privacy policies that consumers could control location collection through the device’s location settings. [1] It alleged those developers had no reason to know the company tracked location regardless of those settings. [1]
The host’s promises become inaccurate
That allegation shows how a third party can make the host’s statements untrue. The host told its users something accurate about its own code. The embedded code did something else. The users relied on the host’s word.
For a bank, a health system or a public agency with a mobile app, the host position is familiar. Apps include analytics, crash reporting, advertising, messaging and mapping components from other firms. Each can collect data under the host’s name and the host’s consent screen.
What the integration guide said
The complaint quoted the company’s integration guides. For Android, the guide told developers to add the coarse and fine location permissions to allow geo-targeted ads. [1] The complaint alleged that providing those permissions was not the only way the company tracked location. [1]
A host’s developers read integration guides to decide what permissions to request and what to disclose. If the guide is incomplete, the host’s disclosures will be incomplete too. In our reading, that is a reason to test what an SDK does as well as read what its guide says.
Testing what leaves the device
Testing can take several forms. A team can inspect network traffic from a test device running the app, with location turned off, and see what data the embedded components send. It can review the permissions each component uses and ask why. It can compare what the app sends with what its privacy notice says.
None of those methods is perfect. Server-side processing, as the complaint described, happens where the host cannot see it. [1] Testing shows what leaves the device, not what the recipient infers from it. Contract terms have to cover the rest.
Contract terms for embedded code
A host can ask its SDK providers for commitments: what data they collect, what they infer, whether they honor device settings, and how they will notify the host of changes. The order itself bound only the company. [2] A host that wants similar protection must write it into its own contracts.
Those commitments matter most for components that can see sensitive data. A banking app’s analytics library may observe account screens. A health app’s messaging component may handle appointment details. A permit app’s mapping component may see a resident’s address. Each is a host relationship with its own risks.
Children in the host’s audience
The complaint alleged that developers had flagged their apps as directed to children through a registration option, and that the company collected personal information from those apps anyway. [1] The 2013 amendments to the COPPA Rule had clarified that the Rule covers an ad network when it has actual knowledge that it is collecting personal information through a child-directed service. [3]
For hosts whose audiences include children, such as a public library, a school district or a pediatric practice, the flag they set with each provider is a statement with consequences. The provider’s knowledge depends on it. The host should confirm that the flag is set and that the provider treats it as the host expects.
Changes the host cannot see
The complaint said the company responded to the investigation in two steps: a new SDK version in November 2015 and server-side changes in December 2015. [1] The first change required app developers to update. The second took effect without any action by them.
That split shows the host’s limited view. A host can control which SDK version it ships, and it can test that version. It cannot see changes on the provider’s servers, which can alter what is done with the data the SDK sends. Contract terms requiring notice of material changes are the only way the host learns of them.
Version control helps with the first kind of change. A host that records which SDK versions are in which app releases can tell, when a problem is reported, which customers ran the affected code. That record also supports any notice the host decides to give its own users.
Whether to carry advertising code at all
Some regulated organizations may decide that their apps should carry no advertising components. That removes one class of embedded code whose business purpose is to use data about the user. It does not remove analytics, crash reporting or other components, which still collect data.
The decision is a business and legal judgment. The case shows the risk on one side: an advertising network whose geo-targeting products, according to the complaint, targeted consumers by current location, by conditional patterns and by location history. [1] A host carrying such a component inherits a share of the questions it raises.
The proof: showing that controls work over time
The fourth part is evidence. The complaint alleged that, despite its stated policies, the company neither implemented adequate privacy controls to ensure COPPA compliance nor tested that the controls it had implemented functioned as intended. [1]
That allegation is about the gap between a control that exists and a control that works. The registration option existed. The complaint alleged that information was still collected from apps that used it. [1]
A privacy program in writing
The order required the company to establish and maintain a comprehensive privacy program, fully documented in writing. [2] The program had to address privacy risks in developing and managing new and existing products, and protect the privacy and confidentiality of covered information. [2]
The order listed elements. They included designating responsible employees, identifying reasonably foreseeable risks, designing and testing controls, selecting and retaining capable service providers and requiring them by contract to protect the information, and adjusting the program as circumstances change. [2]
Independent assessments
The order required an initial assessment and biennial assessments by a qualified, objective, independent third-party professional, covering each two-year period for twenty years after service of the order. [2] Each assessment had to set out the controls in place and certify that the program operated with sufficient effectiveness to provide reasonable assurance. [2]
The assessor needed at least three years of experience in privacy and data protection, and had to be approved by an FTC official. [2] In our reading, that requirement turned the privacy program into something a third party would test and sign, on a schedule, for a long time.
Who signs off inside the organization
The order required the company to designate one or more employees to coordinate and be responsible for the privacy program. [2] A named owner matters because derived data and embedded code cross team lines. Product teams choose SDKs. Data teams build models. Security teams review traffic. Legal teams write notices.
Without one owner, each team can assume another has checked. In our reading, the owner’s first task is a single register that links each app, each embedded component, each derived field and each consent signal. The register is the place where the four parts of this analysis meet.
The owner also needs authority to stop a release. A privacy review that can only advise, after the release date is fixed, will lose to the schedule. Giving the owner a defined veto over releases that add new components or new inferences is a governance choice each organization makes for itself.
Order acknowledgments
The order required the company to deliver copies of the order to principals, officers, directors and managers, and to employees and agents with responsibility for its subject matter, for twenty years. [2] It also required the company to obtain a signed and dated acknowledgment of receipt from each person who received a copy, within thirty days. [2] Delivery spreads knowledge of the obligations inside the company. The acknowledgments create a record that people were told.
What an organization outside the order can take from it
An organization outside the order has no obligations under it. [2] Its structure still offers a model: a written program, named owners, controls tested against what they claim to do, and periodic review by someone independent.
For a regulated institution, those elements will be familiar from other regimes. The useful addition from this case is the focus on derived data and embedded code. A privacy assessment that reviews only first-party collection would miss both, in our reading.
Testing the controls that depend on a flag
The children’s allegations turned on a flag. Developers marked their apps as child-directed, and the complaint alleged the company’s systems still collected and used personal information from those apps. [1] A flag that is recorded but not enforced is a control on paper.
Testing such a control means more than confirming the flag exists. It means sending test traffic from a flagged app and confirming that the data the flag should stop is not collected or used. The complaint’s allegation that controls were not tested to see whether they functioned as intended describes exactly that gap. [1]
What an assessment should cover
The order required each assessment to set out the controls in place, explain why they fit the company’s size and the sensitivity of the data, and explain how they met the order’s requirements. [2] An organization designing its own review can borrow that structure.
For a regulated institution, the review should name derived data and embedded components explicitly. A review scoped only to the organization’s own collection and its own servers would not have caught the conduct the complaint described, in our reading. [1]
What the order did not settle
The order resolved one company’s case. It left several questions for others to answer.
Other inferences. Its definition addressed location only. [2] It did not say how regulators would treat inferences of income, health or other attributes.
Other signals. It named BSSIDs and excluded city-level IP inference. [2] It did not list every signal that can locate a device, such as Bluetooth beacons or cell identifiers. The definition’s general branch, inference from any other data collected through an interface, would carry the weight. [2]
The host’s duty. It bound the company, not the developers whose apps carried its SDK. [2] Whether a host has its own exposure when an embedded component misbehaves depends on the host’s statements and the law that applies to it.
Penalty and conduct. The stipulated judgment was $4,000,000, with the remainder suspended once the required payments were made, on the strength of the company’s sworn financial statements. [2] Because the suspension rested on the company’s finances, the amount payable should not be read as a measure of the conduct alleged, in our reading.
How the issues land in banks, health systems and agencies
The four parts apply wherever an organization ships an app with embedded code. Their weight differs by sector.
Banks and credit unions
Banking apps may use location for fraud detection, branch and ATM finders, or card controls. Each use has a purpose a customer can understand. The risk the case illustrates is a component that uses the same signals for a different purpose, such as advertising, without the customer’s knowledge. [1]
A bank’s fraud team may also want precise location to judge whether a transaction is plausible. The third condition in the order, honoring a consumer’s choice of approximate location, shows the tension. [2] A bank deciding how to use location for fraud should decide with privacy counsel what consent it relies on.
Health systems and digital health
Health system apps handle appointments, messages and records. Location can reveal where a patient receives care. A component that infers location inside a health app may learn something sensitive about the patient’s health.
Health systems that are HIPAA covered entities answer to HIPAA for protected health information. Apps outside HIPAA answer to other regimes, including the FTC’s Health Breach Notification Rule as amended in 2024. [5] Which regime applies to a particular app is a question for counsel.
Public agencies
Agency apps cover benefits, permits, transit and public safety. Residents may have no alternative to the agency’s app for some services. That makes the consent question sharper: a resident who must use the app has less room to refuse a component they do not want.
An agency procuring an app from a vendor is a host in the sense used here. Its contract should state what embedded components may collect and infer, and require notice of changes. That is our suggestion, not a requirement drawn from the case.
Across sectors: the procurement moment
In each sector, the moment of greatest control is procurement. Before an app goes live, the organization can choose its components, require terms and test behavior. After launch, each change competes with other work.
A short standard for embedded code can be applied at procurement and at each major release. It can cover the four parts of this analysis: what is inferred, how consent is read, what the component may do in the host, and what evidence the provider will give. That standard is our suggestion, drawn from the structure of the 2016 order. [2]
Card controls, branch finders and fraud scoring
A bank’s own location features show the range of purposes. A branch finder uses location once, at the customer’s request. A card control that blocks transactions outside a region uses it continuously. A fraud score may compare a device’s location with a transaction’s.
Each purpose implies a different consent and a different precision. A design that collects precise location continuously for all three, because one needs it, runs against the order’s logic of honoring the consumer’s chosen accuracy. [2] Mapping each feature to the least precise location it needs is a practical start.
Patient apps and sensitive places
For a health system, the place itself can be health information. Repeated visits to an oncology center or a behavioral health clinic say something about the visitor. An embedded component that infers those visits learns something the patient may not have shared with anyone.
That is a reason for health systems to restrict location use in patient apps to clear clinical or service purposes, and to test embedded components with particular care. Whether a specific use is permitted under HIPAA or another regime is a question for counsel.
Residents who cannot opt out
Agencies face a further constraint. A resident applying for benefits or reporting to a court program may be required to use an agency app. Consent given under that requirement is weaker than consent to a commercial app the resident could delete.
An agency can respond by limiting what its app collects to what the service needs, and by excluding components whose purpose is unrelated to the service. Its records and privacy obligations, which differ by level of government, will shape the details.
Objections to this reading
Four objections deserve an answer. The first three are numbered below, and the fourth follows them.
- The case was about deception, not inference. The counts charged misrepresentations and COPPA violations. [1] The definition of location information and the consent conditions are in the order’s remedy. [2] Our argument rests on the remedy. In our view, remedies show how a regulator understood the harm, and this one named inference explicitly.
- The order binds one company. It does. [2] We do not claim it binds anyone else. We claim its structure is a usable model for organizations deciding how to govern derived data and embedded code.
- Operating systems have since tightened controls. Platform changes can make some techniques harder. They do not remove the host’s need to know what embedded code does, and they do not reach server-side inference, which the complaint described. [1]
A fourth objection is that inference is unavoidable. Any service that sees an IP address learns roughly where a user is. The order accepted that, by excluding city-level IP inference from its definition. [2] Our argument concerns precise inference that overrides a stated choice, which is what the order’s consent conditions addressed. [2]
Checks a technology or privacy leader can run now
These checks follow from the case’s structure. They are good practice, not requirements stated in this form, and each organization should confirm its obligations with counsel.
Inference
- List every field in your data stores that holds a location, including fields derived from IP addresses, Wi-Fi or other signals, and record its precision.
- Identify models or rules that infer a sensitive attribute from other data, and classify the output at the sensitivity of the attribute it estimates.
Consent
- Confirm that each location-related call or inference in your apps is gated by a current check of the device permission and setting.
- Confirm that a consumer’s choice of approximate location is honored by every component, including server-side processing.
Embedded code
- Inventory the third-party SDKs in each app and record what data each collects and why.
- Test the app with location turned off and record what data leaves the device.
- Review SDK contracts for commitments on data use, inference and notice of changes.
Evidence
- Keep logs showing consent checks before collection, and keep them long enough to answer questions later.
- Schedule periodic independent review of the privacy program, covering derived data and embedded components.
With hindsight
The children’s privacy rule the case relied on was amended again. On April 22, 2025, the FTC published amendments to the COPPA Rule, effective June 23, 2025. [4] Except for certain provisions, regulated entities had until April 22, 2026 to comply. [4]
The 2025 rulemaking addressed the kind of disclosure at issue in 2016. The Commission said the amendments clarify that operators must obtain separate verifiable parental consent for disclosures to third parties. [4] It gave ad networks as an example of such third parties. [4]
The amendments also expanded definitions. Personal information now includes government-issued identifiers and biometric identifiers that can be used for automated or semi-automated recognition. [4] Online contact information now includes mobile telephone numbers. [4]
The amended rule added governance duties. Operators must establish, implement and maintain a written information security program with safeguards appropriate to the sensitivity of children’s information and the operator’s size and complexity. [4] They must keep a written data retention policy specifying their business need for retaining children’s information and their timeframe for deleting it. [4]
On geolocation, some commenters asked the Commission to add geolocation data and other sensitive categories to the definition of personal information. The Commission noted that the definition already includes geolocation data sufficient to identify street name and name of a city or town. [4] The 2013 amendments had added geolocation information to that definition. [3]
Location also mattered in health apps. On May 30, 2024, the FTC published amendments to its Health Breach Notification Rule, effective July 29, 2024. [5] The rule applies to vendors of personal health records and related entities not covered by HIPAA. [5]
The amendments clarified the rule’s coverage of developers of many health apps. [5] They revised the definition of breach of security to include unauthorized disclosures, as well as data security breaches. [5]
One of the rule’s examples involves a health advice app, not covered by HIPAA, that collects geolocation data through a programming interface. [5] The Commission said that the app could draw information from multiple sources, consumer inputs and geolocation, and was therefore a personal health record. [5] In our reading, collecting location through a programming interface, the route the 2016 order’s definition named first, was one of the facts that brought the example app within a breach notification rule.
The amended rule also discussed a Commission complaint against a fertility app developer that, as the rule described the complaint’s allegations, shared information with advertisers and others through SDKs embedded in the app. [5] That case concerned health data rather than location. It shows the embedded-code problem the 2016 complaint described, appearing in a health setting. [1] [5]
The 2016 order itself remains a record of how the Commission framed the remedy at the time. We did not find, in the sources we checked, any later modification of its terms, and we do not report on the company’s compliance with it. [2]
So what held up? In our reading, the 2016 order’s two main ideas, that inferred location is location and that device settings bind embedded code, fit the later rules’ direction. [2] [4] [5] The later rules did not adopt the order’s definition word for word, and we do not claim they did.
What changed is the setting. The 2025 COPPA amendments put third-party disclosures, written security programs and retention policies into the rule itself for operators covered by it. [4] The 2024 health rule reached apps outside HIPAA and counted unauthorized disclosures as breaches. [5] Organizations should check which of these rules applies to each of their apps with counsel.
Spotted an error? Tell us and we will check it and correct the article where needed.