Hide My Email is a feature Apple sells as part of iCloud+, its paid subscription tier. It generates a random address, usually two words and a number ending in icloud.com, that a person can give to a website, a newsletter, or a stranger in place of their real address, with mail sent to the alias forwarded to the real inbox behind it. Apple’s own documentation describes the purpose plainly, telling subscribers they can use the feature to keep your personal email address private so that they do not have to hand over their real address when filling out a form, signing up for a newsletter, or sending mail.¹ When the feature is used through Sign in with Apple, the account-creation system built into many apps and websites, Apple’s documentation goes further, stating that only the app or website an account was created with can use that address to reach the user, and that the personal address stays private.²

The value of the feature rests entirely on that concealment holding. A person uses an alias precisely so that the party on the other end of it cannot learn, keep, or later leak the real address, whether the aim is to cut spam, to keep an account from being tied to their identity, or to limit the damage if the service is later breached.

What the researcher found

On July 1, 2026, the independent technology outlet 404 Media reported that Hide My Email contains a flaw that lets almost anyone take one of these aliases and recover the real address hidden behind it.³ The outlet did not rely on the researcher’s account alone. A 404 Media reporter generated a new Hide My Email address, an alias that had never been used, and passed it to the researcher, who returned the real email address tied to the reporter’s Apple account about five minutes later.³ Digital Trends, checking the same claim, described the same outcome.⁴

The researcher is Tyler Murphy, co-founder of EasyOptOuts, a service that removes people’s records from data brokers.³ In his own testing with volunteers, every alias he checked could be undone, though he was careful to say that the full scope is not known and that his sample was limited.³ What the reporting establishes about the attack, without describing how it works, is that it is fast, that it returned the correct answer each time it was run, and that its only starting material is the alias itself, the string the feature exists to make safe to share.

It helps to place this against the attacks a reader is more likely to have heard of. The familiar routes into an email account or the accounts attached to it, phishing and its telephone and text-message forms, credential stuffing, password spraying, and the various forms of brute-force password guessing, including dictionary and rainbow-table methods, share one feature, which is that each depends on the target doing something wrong, surrendering a password, reusing one across services, or entering credentials where they should not.⁵ What 404 Media reported requires none of that. The person whose address is exposed need not click anything, reuse anything, or make any mistake, because the only thing the attack consumes is the alias the feature already told them was safe to give away.

Neither the researcher nor the outlet published the method, and 404 Media said it was withholding the technical details because the flaw could still be exploited at the time it went to press.³ A method whose finders decline to release it because releasing it would put users at risk is one whose danger lies in how easily it can be followed once known, rather than in any special difficulty of following it.

The timeline

Murphy reported the flaw to Apple in June 2025, and gave the company instructions for reproducing it.³ Apple replied about a month later and said it was looking into the matter.³ In March 2026, Apple told him it had addressed the problem in a recent change to its systems, language that ordinarily signals a fix.⁴ Murphy tested the feature again, found the address still exposed, and sent Apple more information.³ Through the spring, Apple continued to tell him it was investigating, and asked him not to make the issue public until its work was finished.³ At the end of May 2026, Apple said a fix was expected in a security update in the coming weeks.³ That update had not arrived when Murphy, saying he no longer felt comfortable waiting, brought the issue to 404 Media, more than a year after he first reported it, and Apple did not respond to the outlet’s requests for comment.³

Apple publishes its own account of how it handles reports like Murphy’s. On its security research page, the company states that it resolves most reports within ninety days, and that it does not discuss a security issue publicly until its investigation is complete and any needed update has reached users.⁶ Murphy’s report was acknowledged, described at one point as addressed, and then left open past a year, while the feature it concerns remained on sale.

Figure 1. Apple’s handling of the report, set against its own stated ninety-day resolution target.

Why an exposed address matters

The reason an exposed email address is a question of safety rather than nuisance is that an address is a durable key into a large and well-documented market in personal information. In 2014 the Federal Trade Commission published a study of nine data brokers, companies that assemble profiles of individuals from public records, purchases, and online activity and sell them on, generally without the knowledge of the people described.⁷ The Commission found that these firms hold billions of records covering nearly every American, with a single broker in the study holding information on more than 1.4 billion consumer transactions and hundreds of billions of individual data points.⁸ One of the nine, the aggregator Rapleaf, held at least one data point tied to more than eighty percent of all United States consumer email addresses, and attached to those addresses details such as age, marital status, and dozens of further attributes.⁷

The address does not by itself create any of this information. It connects to it. Almost every adult already sits in these databases in the ordinary course of life, through property records, voter rolls, court filings, purchases, and public social accounts, and that profile exists whether or not anyone holds the person’s email.⁷ What the real address supplies is the link between a particular alias and that existing file, the point at which an entry that reads only as a random string of characters becomes attached to a name the database already held. Recovering the address behind an alias is in that sense less the disclosure of a dossier than the bridge to one already assembled.

Figure 2. The recovered address does not build the file; it connects the alias to one the brokers already hold.

A particular kind of data broker, the people-search site, sits at the retail end of this market. The FTC describes these sites as businesses that gather personal information from public records, social media, and other brokers, compile it into a report, and sell that report to anyone willing to pay.⁹ There is, the Commission notes, no federal law that prevents a company from publishing this information, and the burden of removal falls on the individual, one site at a time.¹⁰ For most people the result is unwanted exposure, and for some, including survivors of stalking and domestic violence, the exposure is dangerous enough that a number of states maintain address-confidentiality programs to keep their whereabouts out of these databases.¹⁰ Murphy pointed directly at this market when he noted that free, publicly accessible people-search sites make it simple to turn an address into a name and more, which is why, in his account, the people most likely to rely on Hide My Email for their safety are the ones the flaw most exposes.³

What Hide My Email does not conceal

The concealment Apple offers has a second limit, one that is not a flaw and was never promised away, but that sits at a distance from the sense of anonymity the feature’s name invites. Hide My Email hides the real address from the app, the site, or the sender that receives an alias. It does not hide the connection from Apple, which holds the mapping between every alias and the account behind it, and produces that mapping in response to lawful demands.

In early 2026, court records reported by 404 Media and TechCrunch showed how this works.^{11, 12} Federal agents investigating a threatening message sent to Alexis Wilkins, identified in filings as the girlfriend of FBI Director Kash Patel, traced the message to a Hide My Email alias.¹¹ Apple, served with a legal request, provided the account holder’s real name and address, identifying the sender, and turned over records for the 134 aliases that account had generated.^{12, 13} TechCrunch reviewed a second warrant in which Apple provided similar records to Homeland Security Investigations during an identity-fraud inquiry.¹² In each case the feature worked as designed against the third parties who received the aliases, and provided no barrier at all to a subpoena, because it was never built to.

The change now underway

While the flaw sat open, Apple prepared a separate change to the feature that will reduce its usefulness for reasons unrelated to the vulnerability. In June 2026, TechCrunch reported that Apple intends to move newly generated aliases from the icloud.com domain to a distinct private.icloud.com domain, and to shift the addresses created through Sign in with Apple away from the privaterelay.appleid.com domain used today.^{14, 15} An address that announces on its face that it is a relay is one that a website or a mailing list can identify and refuse, which would leave users of the feature to supply a real address after all.¹⁵ The stated purpose of the change is to help email systems prepare for the switch, and its effect is to make the aliases easier to block at the point where they are meant to protect the user.¹⁵

What a reader can do

For a reader who uses Hide My Email, the practical meaning of the flaw follows from what an exposed address does. The immediate and visible effect of recovery may be no more than additional unwanted mail reaching the real inbox. The consequential effect, and the one that matters for the people the feature is meant to protect, is that an identifier the reader had kept concealed becomes a known one, and a known address is what allows the alias to be tied through the data-broker and people-search market to a name and a location, in the way set out above.^{7, 9} The full extent of who is affected is not established, because the researcher’s testing was limited and the method is not public.³

There are steps within a reader’s own control, though it should be said plainly that none of them closes the flaw, because the flaw is Apple’s to fix and no setting available to a user removes it. A person especially concerned about a particular alias can deactivate it in the Hide My Email settings, after which it no longer forwards mail, and can generate a new address for the service in question, though deactivating an alias does not change the real address behind it and so does not undo a recovery that has already occurred.¹ The more durable protection is against the downstream harm rather than the flaw itself, because the danger of a recovered address lies in what the people-search market can attach to it, so a reader can reduce that exposure by opting out of those sites, a process the Federal Trade Commission sets out, which requires a separate request to each site and repetition over time as records reappear.⁹ And because an alias no longer carries the assurance the feature’s name implies, the concealment cannot be relied on when deciding whether to share one with a party a reader does not trust, until Apple ships the fix it said in late May was expected in the coming weeks.³

The record

The record now stands as follows. Apple sells a feature whose stated purpose is to keep a person’s real email address private from the parties they share an alias with, and charges a subscription for it. A researcher found that the address can be recovered from the alias by almost anyone, in about five minutes, with the alias as the only input, reported it to Apple with reproduction steps in June 2025, was told in March 2026 that it had been addressed, found that it had not, and watched the feature remain both exploitable and on sale more than a year later. Separately, the same feature has been shown to yield its users’ real identities to law enforcement on request, and is being changed in a way that will make its aliases easier for websites to reject. Each of these is documented, and each is left for the reader to weigh.

End Notes

1. Apple Support, “Create and manage Hide My Email addresses in Settings on iPhone.” https://support.apple.com/guide/iphone/create-and-manage-hide-my-email-addresses-iphcb02e76f7/ios

2. Apple Support, “How to use Hide My Email with Sign in with Apple.” https://support.apple.com/en-au/105078

3. Joseph Cox, “Apple ‘Hide My Email’ Vulnerability Reveals Peoples’ Real Email Addresses,” 404 Media, July 1, 2026. https://www.404media.co/apple-hide-my-email-vulnerability-reveals-peoples-real-email-addresses/

4. “Apple’s Hide My Email feature has an unfixed bug that leaves email addresses exposed,” Digital Trends, July 2026. https://www.digitaltrends.com/computing/apples-hide-my-email-feature-has-an-unfixed-bug-that-leaves-email-addresses-exposed/

5. For these attack categories, see Cybersecurity and Infrastructure Security Agency, National Initiative for Cybersecurity Careers and Studies, “Password Spraying and Credential Stuffing.” https://niccs.cisa.gov/training/catalog/cybrary/password-spraying-and-credential-stuffing

6. Apple Support, “Report a security or privacy vulnerability.” https://support.apple.com/en-us/102549

7. Federal Trade Commission, “Data Brokers: A Call for Transparency and Accountability,” May 2014. https://www.ftc.gov/system/files/documents/reports/data-brokers-call-transparency-accountability-report-federal-trade-commission-may-2014/140527databrokerreport.pdf

8. Federal Trade Commission, “FTC Recommends Congress Require the Data Broker Industry to be More Transparent,” May 27, 2014. https://www.ftc.gov/news-events/news/press-releases/2014/05/ftc-recommends-congress-require-data-broker-industry-be-more-transparent-give-consumers-greater

9. Federal Trade Commission, “What To Know About People Search Sites That Sell Your Information.” https://consumer.ftc.gov/articles/what-know-about-people-search-sites-sell-your-information

10. Privacy Rights Clearinghouse, “FAQ: Online Data Brokers and People Search Sites.” https://privacyrights.org/resources-tools/archives/faq-online-data-brokers-people-search-sites

11. Joseph Cox, “Apple Gives FBI a User’s Real Name Hidden Behind ‘Hide My Email’ Feature,” 404 Media. https://www.404media.co/apple-gives-fbi-a-users-real-name-hidden-behind-hide-my-email-feature/

12. “Apple will hide your email address from apps and websites, but not cops,” TechCrunch, March 30, 2026. https://techcrunch.com/2026/03/30/apple-will-hide-your-email-address-from-apps-and-websites-but-not-cops/

13. “Apple revealed ‘Hide My Email’ user identity in FBI investigation,” CyberInsider, March 2026. https://cyberinsider.com/apple-revealed-hide-my-email-user-identity-in-fbi-investigation/

14. “Apple plans to change its Hide My Email privacy feature that could make it less effective,” TechCrunch, June 16, 2026. https://techcrunch.com/2026/06/16/apple-plans-to-change-its-hide-my-email-privacy-feature-that-could-make-it-less-effective/

15. “Apple hasn’t fixed a Hide My Email privacy bug in over a year,” AppleInsider, July 1, 2026. https://appleinsider.com/articles/26/07/01/apple-hasnt-fixed-a-hide-my-email-privacy-bug-in-over-a-year

Figure 1. A satellite receiver takes in signals and computes the phone’s position on its own. The result leaves the phone only when software on the phone transmits it.

The connection itself comes first. For a phone to receive a call or a message, the network has to know where to send it, so the phone continually announces its presence to the nearest cell towers and announces it again as the person moves. The carrier therefore holds, at every moment, a record of which tower and which sector is serving each phone, and from the timing and strength of the signal that record can be narrowed considerably. The rules the government wrote for emergency calls require carriers to locate a caller closely enough to identify the building in most cases, which is the level of precision the network is capable of producing when it is asked to.¹ None of this depends on the phone’s satellite receiver, and none of it can be switched off by a person who wishes to remain reachable. The record exists because the phone works the way it must in order to be a phone.

Figure 2. To remain reachable, a phone keeps registering with the nearest towers, so the carrier holds a continuous record of which cell is serving it, narrowed by timing and signal strength.

That record is the subject of the one clear rule in this area. The Supreme Court has held that when the government obtains the accumulated cell-tower records that trace a person’s movements over an extended period, it conducts a search and needs a warrant.² The same data, however, has moved through commerce with far less friction. Until 2019 the major carriers sold access to their customers’ location information to intermediaries, who resold it to location-finding services, and in at least one documented case a sheriff obtained a person’s location after uploading, in place of a warrant, paperwork that had nothing to do with the request. The Federal Communications Commission fined the four national carriers a combined total near two hundred million dollars, on the ground that a phone’s location is customer information the carriers were required to keep confidential and had not. The carriers contested the penalties on procedural grounds for years, and in 2026 the Supreme Court resolved that procedural dispute in the government’s favor, leaving the penalties in place.³

The second route does not wait for the carrier. A cell-site simulator, known in law enforcement for years by the trade name Stingray, is a portable device that broadcasts as though it were a legitimate cell tower, and broadcasts more strongly than the real towers nearby, so that every phone within range treats it as the most attractive tower and registers with it. In doing so each phone gives up the identifiers that distinguish it, and the device measures the phone’s signal and fixes its location, through the walls of a home or an office.⁴ Because it imitates the network rather than photographing anything, it needs no clear line of sight, and because it compels every phone in the area to answer, it collects the phones of bystanders who are the target of nothing.

Figure 3. A cell-site simulator broadcasts as a stronger tower so that nearby phones register with it, revealing their identifiers and location, including the phones of bystanders.

How that device has been governed is seen most clearly in New York, where the record is unusually complete because a civil-liberties organization pried it loose. In response to a public-records request, the New York Police Department disclosed that it had used cell-site simulators more than a thousand times between 2008 and 2015 with no written policy in place, and that its practice had been to obtain not a warrant but a pen register order, a lower form of court authorization meant for recording the numbers a telephone dials.⁵ In 2017 a Brooklyn court, in People v. Gordon, held that a pen register order does not authorize a cell-site simulator at all, and that using one to find a person is a search requiring a warrant supported by probable cause.⁶ The Police Department has since adopted a policy that requires a warrant in most cases, but it did so under a city disclosure law and a court ruling rather than a state statute.⁷ New York itself has enacted nothing. Bills to require a warrant, and bills to bar the devices outright, have been introduced repeatedly and have not passed, and while the federal government requires its own agents to obtain warrants as a matter of Justice Department policy, and a handful of states have written the requirement into law, the requirement remains a patchwork rather than a rule.⁸

The third route reaches a phone without touching it and without the carrier’s cooperation, by exploiting the system the carriers use to speak to one another. To connect a call that crosses from one network to another, and to let a phone work while it roams, the world’s carriers exchange messages over a set of signaling protocols, the oldest of which was designed in the nineteen-seventies for a small circle of national telephone companies that had no reason to distrust one another. That trust is now the weakness. A party with access to the signaling network can ask it, in effect, which cell is serving a given telephone number, and receive an answer, from anywhere in the world, while the phone displays nothing and no software is installed on it. The same access can be used to intercept calls and text messages and, in some configurations, to deliver spyware to the device.⁹

Figure 4. A party with access to the signaling network that carriers share can request a phone’s serving cell from anywhere, with nothing shown on the device.

This is not a theoretical weakness. In 2024 a senior official at the federal cybersecurity agency told the Communications Commission, in a public filing, that he had seen reliable information of numerous attempts to track people inside the United States through these protocols, and that the reported incidents were likely a small fraction of the total.¹⁰ A cybersecurity firm reported that as recently as late 2024 it had watched a surveillance vendor exploit a newly discovered method of getting around the defenses the carriers had installed.¹¹ Because the party doing the tracking is frequently outside the country, no American warrant reaches it, and no American statute governs it.

The fourth route is the one place where the satellite receiver does the damage, and it does so only because the phone’s owner has been induced to give the reading away. Thousands of ordinary applications, a weather app, a game, a coupon service, ask permission to know the device’s location and then pass the precise satellite fix to the commercial data market, either through code the app is paid to include or through the automated auctions that place advertisements.¹² Brokers assemble those readings into histories of individual movement and sell them.

Turning off what the phone calls “location” closes one route and leaves three open.

It stops the applications from reading the satellite fix, but the phone still registers with the towers, still answers a simulator, and can still be found through the signaling network, because none of those depend on a setting the owner controls.¹⁹

Another, together with its parent company, told regulators it handled more than seventeen billion location signals from roughly a billion phones every day, and that the data it sold was not anonymized and could be traced to individuals.¹⁴ The Federal Trade Commission has brought a series of actions against these firms, barring several of them from selling location data that reveals visits to sensitive places such as medical clinics and houses of worship, but the orders carve out exceptions that include national security and law enforcement, which is the use that matters here.¹⁵

What turns that market into a route around the law is that agencies buy from it.

In 2026 the director of the Federal Bureau of Investigation confirmed to a Senate committee that the Bureau buys commercially available data capable of tracking Americans’ movements, and, asked whether it would stop, made no commitment.¹⁷ Legislation to close the gap, by forbidding the government to purchase what it would otherwise need a warrant to obtain, passed the House of Representatives in 2024 and was not taken up by the Senate.¹⁸

Set side by side, the four routes describe the same fact about the same person, where they are and where they have been, reached by four different means and governed by four different standards. The carrier’s own record is protected by a warrant requirement the Supreme Court has recognized.

The impersonator that summons a phone from the street is governed by an agency policy and by the laws of a few states, and by no federal statute. The signaling network that locates a phone from abroad is governed by nothing that can reach the party using it. The purchased fix is restrained only by consent orders that exempt the government. The one component a person is most likely to picture as the threat, the satellite receiver, is the one that reports nothing on its own.

Figure 5. The same location, reached four ways. A warrant is required for the carrier’s record and, in practice, for none of the others.

None of these routes is new, which is what makes the absence of law around them a finding rather than a delay. The cellular network has located phones for as long as phones have been mobile, and each way of reading it has been documented for years, in court records, in regulatory filings, and in the reports of the agencies themselves. What has not arrived is the law.

New York, which could require a warrant for the impersonator and has been asked to for a decade, has not. Congress, which could close the market in purchased location and has voted on the question, has not. The signaling network has no owner willing to answer for it. In each instance the gap is not an oversight waiting to be corrected but the settled condition under which the tracking proceeds.

The result is a device that reports a person’s location at nearly every layer of its operation. It registers with the towers to stay reachable, and the registration is a record. It answers a stronger signal, and the answer can be forced by a box on the street.

And the location it computes for its own use can be bought on the open market from the applications its owner installed.

The one part that stays silent is the satellite receiver most people would name first. A person can switch off very little of this and still carry a working phone, which is the point. The exposure is not a feature that was added to the phone. It is the way the phone connects.

End Notes

1. The 50-meter horizontal location accuracy metric adopted for wireless 911 calls, sufficient to identify the building for most calls, appears in the Commission’s Fourth Report and Order (2015); see the FCC’s overview of 911 and E911 services, https://www.fcc.gov/general/9-1-1-and-e9-1-1-services.

2. Carpenter v. United States, 585 U.S. 296 (2018).

3. Federal Communications Commission v. AT&T, Inc., 608 U.S. ___ (2026), which describes the carriers’ sale of customer location data through intermediaries, the incident in which a sheriff obtained location data after uploading irrelevant paperwork, and the penalties imposed under the customer-information confidentiality provision, and which held that the Commission’s forfeiture procedure was constitutional, https://www.supremecourt.gov/opinions/25pdf/25-406_nmip.pdf. The penalties were approximately 57 million dollars against AT&T, 47 million against Verizon, and roughly 92 million against T-Mobile and the former Sprint.

4. On the operation of cell-site simulators, see the New York Civil Liberties Union memorandum in support of a warrant requirement, https://www.nyclu.org/uploads/2015/04/memo_stingrayuse_NY_201508_final.pdf.

5. New York Civil Liberties Union, NYPD Has Used Stingrays More Than 1,000 Times Since 2008, reporting 1,016 uses between 2008 and May 2015 and the absence of a written policy, https://www.nyclu.org/press-release/nypd-has-used-stingrays-more-1000-times-2008.

6. People v. Gordon, 2017 NY Slip Op 27364 (Sup. Ct., Kings County).

7. New York Police Department, Cell-Site Simulators: Impact and Use Policy (Nov. 24, 2023), which states that in most cases investigators must first obtain a search warrant, https://www.nyc.gov/assets/nypd/downloads/pdf/public_information/post-final/cell-site-simulators-nypd-Impact-and-use-policy_11.24.23.pdf.

8. For a pending New York warrant bill, see Senate Bill S6567 (2025-2026), https://www.nysenate.gov/legislation/bills/2025/S6567. On the Justice Department’s 2015 warrant policy and the states that have enacted warrant requirements, see Project On Government Oversight, Issue Brief: The Cell-Site Simulator Warrant Act, https://www.pogo.org/fact-sheets/issue-brief-the-cell-site-simulator-warrant-act.

9. On the vulnerabilities of the SS7 signaling network, see Electronic Frontier Foundation, EFF to FCC: SS7 is Vulnerable, and Telecoms Must Acknowledge That, https://www.eff.org/deeplinks/2024/07/eff-fcc-ss7-vulnerable-and-telecoms-must-acknowledge.

10. The statement of the Cybersecurity and Infrastructure Security Agency official, Kevin Briggs, to the Commission is described in CISA Official Warns US Telecom Networks Are Vulnerable to Spying, https://www.itbrew.com/stories/2024/05/29/cisa-official-warns-us-telecom-networks-are-vulnerable-to-spying.

11. A Surveillance Vendor Was Caught Exploiting a New SS7 Attack to Track People’s Phone Locations, https://techcrunch.com/2025/07/18/a-surveillance-vendor-was-caught-exploiting-a-new-ss7-attack-to-track-peoples-phone-locations/.

12. On how applications transmit location to the data market through embedded code and real-time bidding, see Electronic Frontier Foundation, Federal Regulators Limit Location Brokers from Selling Your Whereabouts, https://www.eff.org/deeplinks/2024/12/federal-regulators-limit-location-brokers-selling-your-whereabouts-2024-review.

13. On Fog Data Science, its claimed access to data from 250 million devices, its pricing to local police, and its sale of movement data without a warrant, see Electronic Frontier Foundation, Inside Fog Data Science, https://www.eff.org/deeplinks/2022/08/inside-fog-data-science-secretive-company-selling-mass-surveillance-local-police.

14. Federal Trade Commission, FTC Takes Action Against Gravy Analytics, Venntel for Unlawfully Selling Location Data (Dec. 2024), noting processing of over 17 billion signals from roughly a billion devices daily and that the data was not anonymized, https://www.ftc.gov/news-events/news/press-releases/2024/12/ftc-takes-action-against-gravy-analytics-venntel-unlawfully-selling-location-data-tracking-consumers.

15. Federal Trade Commission, FTC Finalizes Order Prohibiting Gravy Analytics, Venntel from Selling Sensitive Location Data (Jan. 2025), https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-finalizes-order-prohibiting-gravy-analytics-venntel-selling-sensitive-location-data. On the national-security and law-enforcement exceptions in these orders, see the Electronic Frontier Foundation review cited above.

16. On the ODNI Senior Advisory Group report and the Defense Intelligence Agency’s written statement that it does not construe Carpenter to require a warrant for purchased commercially available data, see Electronic Privacy Information Center, ODNI Report on Intelligence Agencies’ Data Purchases Underscores Urgency of Reform, https://epic.org/odni-report-on-intelligence-agencies-data-purchases-underscores-urgency-of-reform/.

17. The testimony of FBI Director Kash Patel before the Senate Intelligence Committee, and his refusal to commit to ending the practice, are recounted in Warren Davidson, Americans’ Fourth Amendment Rights Are Not for Sale, https://davidson.house.gov/2026/4/op-ed-americans-fourth-amendment-rights-are-not-for-sale.

18. Fourth Amendment Is Not for Sale Act, H.R. 4639, 118th Congress (2023-2024), https://www.congress.gov/bill/118th-congress/house-bill/4639. The House passed the measure in April 2024; the Senate did not take it up.

19. On the distinction between the location setting, which governs application access to the satellite receiver, and the cellular network's independent record of the serving cell, see Envista Forensics, Is My Cell Phone Tracking Me If I Turn Off Location Services?, https://www.envistaforensics.com/knowledge-center/insights/articles/is-my-cell-phone-tracking-me-if-i-turn-off-location-services/.

A new sensor pairs roadside plate cameras with the radio signals your phone, your earbuds, your watch, and your car itself give off, combining them into a signature that identifies the person rather than the plate and keeping it for later search. No federal law prohibits it, and New York has none of its own.

A company that sells license plate cameras to police departments illustrates its newest product with a hypothetical stretch of road. Of one hundred cars that pass a sensor, about seventy might carry an Apple phone, a fact that on its own distinguishes no one. Only one of the hundred will carry that phone in the particular company it keeps, a phone of a specific model traveling with a specific car radio, a specific pair of wireless headphones, a fitness watch, a small electronic key finder, and the plate ABC-1234. That combination is unique, the company explains, and once a roadside sensor has recorded it the same car can be recognized again by the combination alone, even after the license plate has been removed, covered, or replaced.¹

The product is called SignalTrace. It is made by Leonardo, an Italian aerospace and defense company whose largest shareholder is the Italian government, and it is sold in the United States through the company’s subsidiary, Leonardo US Cyber and Security Solutions.² SignalTrace is not itself a camera. It is a set of radio sensors designed to be mounted alongside the automatic license plate readers that already stand on utility poles, highway overpasses, and patrol cars throughout the country, and its function is to record the wireless identifiers that consumer electronics give off as they travel, then attach those identifiers to the plate and to the time and place a camera recorded it.³

An automatic license plate reader, commonly abbreviated as ALPR, is a camera that photographs the plate of each passing vehicle, converts the image into text, and stores the location and the moment of the sighting. Police use these systems to recover stolen vehicles and to trace cars linked to crimes, and civil liberties organizations have argued for years that the same systems, operated continuously across a city, accumulate a record of where ordinary people drive, when they drive there, and whom they meet.⁴

The argument that has kept the cameras lawful without a warrant, in most courts that have examined them, is that a camera on a public street observes only what any bystander could observe, namely a plate, and that a plate identifies a vehicle rather than the person inside it.⁵ SignalTrace is designed to dissolve the distinction on which that argument rests.

The method, described in the company’s product materials and in the patent it received in 2024, is correlation across time.⁶ Phones, earbuds, watches, laptops, key fobs, and the electronics built into vehicles all transmit brief radio messages in the course of ordinary operation, and many of those messages contain a number that identifies the individual device. What a sensor captures is that identifier, the number the device announces to anything within range, not the location the device works out for itself and not the contents of anything it sends.

The location comes from the sensor rather than the device, because a sensor sits at a known place, so reading an identifier there records where the device, and whoever is carrying it, was at that moment, without the phone computing or transmitting anything.

A single such number, captured once, reveals little. SignalTrace records which devices appear together, repeatedly, in the presence of the same vehicle, and treats that recurring set as a signature for the vehicle and, in turn, for whoever travels with it.

Where a plate camera occupies the same site as a SignalTrace sensor, the system links the signature to the plate. Where no camera is present, the company states, the signature on its own still establishes a location, a date, and a time. The information is retained on the company’s analysis platform so that it can be searched afterward, which means an agency can ask not only where a given plate has been but where a given cluster of devices has traveled, and which other devices have traveled beside it.⁷ The premise that a small set of identifiers can single out one person has independent support in the research literature. A widely cited study of human movement data found that four approximate locations were enough to identify almost every individual in a large anonymized set, and that the result held even as the data was made deliberately coarse.⁸

Figure 1. The system records the set of device identifiers that recur in the presence of a vehicle and ties that recurring set to the plate and to a time and place. The identifier is the combination, not any single device.

Figure 2. Each identifier added narrows the set of possible vehicles. In a large study of location data, four points were enough to single out almost everyone, the same dynamic that makes a small device combination uniquely identifying.

The system reads passively, out of the open air, and a person nearby has no practical means of knowing that a sensor is present or of declining to be read. What such a person might otherwise rely on for protection comes down to a few assumptions about the devices they carry and the car they drive.

One assumption is that a car, unlike a phone, broadcasts nothing that singles it out, when in fact nearly every passenger vehicle on the road does. Federal safety regulations have for years required new passenger cars to carry direct tire pressure sensors, one inside each wheel, and the requirement has been in force long enough that essentially every car now on the road has them.⁹ Each sensor transmits a fixed identification number that is not encrypted, continuously, while the car is in motion.

The standard exists for safety, to warn a driver of an underinflated tire, and tracking was not its purpose, but the effect has been documented for years. Researchers showed more than a decade ago that the number a tire sensor broadcasts is long enough to be unique to a car and can be read from a passing vehicle at a distance of tens of meters with inexpensive equipment, and a peer-reviewed study published in 2026 tracked vehicles across a neighborhood for weeks using receivers assembled from roughly hundred-dollar parts.¹⁰ A driver cannot switch the sensors off, and they broadcast whether or not anyone in the car is carrying a phone. The car was never anonymous; the identifier was installed at the factory to satisfy a federal rule.

A second assumption is that modern phones have already addressed the problem by disguising the identifier they broadcast, and they do, though only in part. To frustrate precisely this kind of tracking, recent versions of both major phone operating systems replace the permanent hardware address a phone would otherwise broadcast over Wi-Fi with a temporary, randomized one.¹¹ The protection is genuine but incomplete. Researchers have shown that the randomization can be defeated on the large majority of devices they tested, by exploiting the timing and sequence of the disguised messages, and that a separate flaw in one manufacturer’s Bluetooth design allows a device to be followed in spite of it.¹²

For SignalTrace the limitation matters less than it might, because the signature the system depends on is the combination of devices rather than any one number, and disguising a single element does not break the set when other elements, the tire sensors most of all, do not change. The attack has also been carried out beyond the laboratory. One research team identified Bluetooth devices in public by the small imperfections in their hardware, which the disguise does not conceal, and followed a willing participant in and out of a home using radios that cost less than two hundred dollars.¹³

Figure 3. Randomization changes the phone’s broadcast identifier from one sighting to the next, but the other identifiers in the cluster, including the federally required tire sensors, do not change. The match is made on the part of the signature that stays constant.

A third assumption is that the published list of what the system can track should be read literally, and at least one item on that list cannot be. Among the device types the company names are the radio-frequency identification tags used in workplace access cards and in pet microchips.¹⁴ A pet microchip contains no battery; it stores no charge and transmits nothing until a reader held within a few centimeters supplies it with power, which is why a veterinarian must pass a scanner directly over the animal to read one.¹⁵ A sensor on a pole reading an implanted chip through the body of a moving car is not something the underlying technology allows. The point is a narrow one, but it is worth stating directly, because it informs how the remainder of the list should be read. The marketing describes capabilities that in at least one case the physics do not permit, and a reader of the document has no straightforward way to tell which of the listed capabilities have been demonstrated and which are aspirations.

Figure 4. The protections a driver is likely to assume, set beside what the relevant hardware actually does.

None of this means the tire sensors reveal anything about the car that the plate does not. As a way to identify a vehicle the two duplicate each other, and a reader who has accepted the plate camera has already surrendered the privacy of the car’s identity on a public road. What the sensors add is not a new fact about the car but a sturdier and cheaper way of collecting the same one. A plate can be removed, covered, exchanged for another, copied onto a second car, or missed by a camera in the dark or at a bad angle, and through all of it the sensors keep sending the same fixed number that the driver cannot switch off. They also read a radio signal rather than an image, so they need no camera, no clear line of sight, and no good light, and they can be made small and cheap enough to place where a camera could not go and to read a car they never see. Whether that is a different kind of surveillance or only a great deal more of the same can be argued either way, but it removes the cost and the sightline that once kept the reach of a camera network in check.

The personal devices are a separate matter, because they do not identify the car. They identify the people in it. A plate is a registration, a record of who a vehicle belongs to rather than who is driving it, whether the owner is present, or who is riding along, and the courts that allowed plate cameras without a warrant did so on the stated ground that a camera reads the vehicle and not the people inside.

A phone, a pair of earbuds, and a watch belong to a person rather than a car, which is the part the reasoning behind the plate was built to leave out. They also make a person harder to lose rather than easier, because the signature is a set rather than a single tag, so leaving your own phone at home does not break the trail when the devices of the people you travel with still place you in the car, and the more regularly you travel with the same people the more ways there are to find you. And much of this layer works where there is no plate, no camera, and no road, on foot, in a station, or in a shopping center, in the places a plate reader cannot follow.

The company anticipates the privacy objection and answers it with a single comparison. Its materials say that the system records the signals devices emit without reading the contents of any communication, and that this is no different from the way a license plate reader records a plate rather than information about the driver. The same materials describe the product’s purpose as the ability to “identify the movements of electronic devices, individuals, and vehicles,”¹⁶ and the executive who announced the underlying patent described the aim as enriching plate data with device information in order “to find the individuals police are looking for.”¹⁷

The reassurance and the description are difficult to reconcile. A product offered to police on the strength of its ability to identify the individual is not, in the same document, a product that records vehicles rather than the people inside them. The company has not resolved the tension; asked about the privacy implications by several news organizations after the product drew public attention, it has not commented beyond its marketing materials.¹⁸

Whether any of this is lawful is, for now, unsettled, and it is unsettled because the law that governs surveillance was written around older facts. The Fourth Amendment forbids unreasonable searches, and for most of its history a search meant a physical intrusion into a place a person could expect to keep private. The Supreme Court has more recently held that the protection can extend to information a person exposes in public, once the government gathers enough of it.

In 2018, in Carpenter v. United States, the Court held that the government conducts a search when it obtains the months of cell-phone location records that trace the whole of a person’s movements, and that the older principle permitting the government to take, without a warrant, any information a person has turned over to a third party does not stretch to so complete a record.¹⁹ Earlier decisions had pointed in the same direction, one Justice noting that society does not expect the police to record and store every movement of a person’s car over a long period, and a federal appeals court concluding that a network of cameras recording almost an entire city’s daily movement amounted to placing an ankle monitor on every resident.²⁰

License plate readers themselves, though, have largely withstood these challenges, and the reasoning that has preserved them is specific. In January 2026 a federal court in Virginia upheld the city of Norfolk’s network of automatic plate cameras against a constitutional challenge brought by two residents whose vehicles had been photographed hundreds of times within a few months, holding that fixed cameras at scattered points capture only occasional, isolated sightings rather than the continuous record condemned in the earlier cases, and that they photograph vehicles rather than identifying the people in them.²¹

Most courts to take up the question have arrived at the same conclusion, and a Virginia trial court that ruled the other way, suppressing plate-camera evidence as the fruit of a warrantless search, was reversed on appeal.²² Those two findings, that the cameras capture only fragments and that they track cars rather than people, are the foundation on which warrantless plate reading has been allowed to rest.

SignalTrace is engineered to remove both. Its stated purpose is to identify the person, and its method, a persistent signature assembled across many sightings, is precisely the continuous and comprehensive record the courts have treated as the boundary a plate camera does not cross. The judge in the Norfolk case acknowledged that the analysis must remain flexible as the technology advances, and that the balance he had struck held, in his phrase, “not today,” for the cameras as they currently function.²³

No statute closes the space the Constitution leaves open. There is no federal law written specifically for license plate readers, and the federal laws that protect electronic privacy were built for communications carried by telephone and internet companies rather than for identifiers broadcast into the air and gathered by a government sensor. The federalwiretap statute protects the contents of a communication, which it defines as information concerning the substance, purport, or meaning of that communication, and it expressly exempts radio communications that are readily accessible to the general public.²⁴

A system that records identifying numbers rather than the substance of any message, and that describes itself as capturing frequencies emitted into the open air, sits outside that statute by design, and the company’s assurance that it does not capture contents is, in practical terms, a statement that the law of wiretapping does not reach it.²⁵ The companion statute that governs the dialing and routing information of telephone calls and internet traffic applies to information transmitted through a provider’s equipment and requires a court order before the government installs such a device, neither of which describes a sensor listening to the open air on its own, and the law protecting stored communications applies to records held by a communications company rather than to data an agency collects and keeps itself.²⁶

The collection therefore falls between the established categories, and the only legal constraint that plainly applies is the Fourth Amendment, whose application to this particular combination no court has yet decided.

Figure 5. The federal privacy statutes and state plate-reader laws each cover a different category. The capture of device identifiers from the open air falls outside all of them, leaving only the Fourth Amendment, whose application is undecided.

New York has done less than the federal government rather than more. The state has no statute restricting how license plate readers may be used, what their data may be used for, how long it may be kept, or with whom it may be shared. This is not an inference but the stated premise of the legislation written to change it.

A bill introduced in the State Senate in April 2026 to create such a framework opens by observing that there is at present no state regulation of the use, transfer, storage, or sharing of plate-reader data by New York agencies or law enforcement.²⁷ What New York has actually enacted is far narrower. A law passed in 2025 directs a state police-training body to develop a model policy on the use of plate readers, a policy that agencies are required to post but that imposes no warrant requirement and creates no enforceable limit, and the law applies, by its own definition, only to systems that convert images of license plates into data.²⁸

The device harvesting that SignalTrace performs is not addressed by that law, nor by the older non-binding guidance the state has issued, nor even by the most far-reaching reform now pending, which would for the first time impose penalties and forbid the use of improperly collected plate data in court, but which speaks only of plates and says nothing of the radio identifiers given off by phones, watches, and cars.²⁹ In New York, then, the part of the system that reads the plate is lightly governed, and the part that reads the person is governed by nothing at all.

The permission that has allowed plate cameras to operate without a warrant was granted on the understanding that they do two things and not two others. They record where a vehicle was at scattered moments, and they read a plate. They do not trace the whole of a person’s movements, and they do not identify the people inside the car. SignalTrace is built to do the two things that the permission was conditioned on its not doing, and it is being offered to the same agencies that already run the cameras, as an addition to equipment they have already purchased. Whether the law will treat the combination as a search has not been decided, and the sensors are being installed while the question waits. They operate passively and without notice, and there is, for the moment, nothing a driver can switch off.

End Notes

1. Leonardo US Cyber and Security Solutions, ELSAG SignalTrace product page (the one-hundred-car illustration and the identification of a vehicle by its device signature without a plate), https://www.leonardocompany-us.com/lpr/elsag-signaltrace

2. The Italian Ministry of Economy and Finance is Leonardo’s largest shareholder, holding roughly 30.2 percent of the company; see Leonardo S.p.A. corporate governance disclosures, https://www.leonardo.com/en/investors/corporate-governance/board-of-directors/composition. The United States subsidiary that markets SignalTrace is Leonardo US Cyber and Security Solutions, LLC, formerly Selex ES.

3. SignalTrace product sheet (2025), Leonardo US Cyber and Security Solutions; first reported by Joseph Cox, 404 Media, This Company Will Add Phone, AirPod, and Smartwatch Trackers to License Plate Readers, https://www.404media.co/this-company-will-add-phone-airpod-and-smartwatch-trackers-to-license-plate-readers/

4. On the longstanding concern that continuous plate reading accumulates a record of drivers’ movements, see Electronic Privacy Information Center, Vehicle Fingerprinting Through Pervasive Camera Surveillance Likely Violates Fourth Amendment, Court Finds, https://epic.org/vehicle-fingerprinting-through-pervasive-camera-surveillance-likely-violates-fourth-amendment-court-finds/

5. On the prevailing view that fixed plate cameras do not require a warrant because a plate is exposed to public view, see Schmidt v. City of Norfolk (E.D. Va. Jan. 27, 2026), reported at https://courthousenews.com/judge-holds-norfolks-license-plate-reader-use-constitutional/

6. United States Patent No. 11,941,716 B2, Systems and Methods for Electronic Signature Tracking, issued March 26, 2024; Leonardo US Cyber and Security Solutions press release, May 30, 2024, https://www.businesswire.com/news/home/20240530567391/en/Leonardo-US-Cyber-and-Security-Solutions-receives-a-patent-for-their-new-electronic-signature-tracking-system-for-law-enforcement

7. SignalTrace product page and product sheet (retention of correlation data for later query; identification of a vehicle by signature with or without a co-located camera), https://www.leonardocompany-us.com/lpr/elsag-signaltrace

8. On the principle that a small set of points uniquely identifies a person, see Yves-Alexandre de Montjoye and others, Unique in the Crowd: The Privacy Bounds of Human Mobility, Scientific Reports volume 3, article 1376 (2013), which found that four approximate time-and-place points identified 95 percent of 1.5 million people in an anonymized data set, https://www.nature.com/articles/srep01376

9. The requirement originates in the Transportation Recall Enhancement, Accountability, and Documentation (TREAD) Act of 2000 and Federal Motor Vehicle Safety Standard No. 138, with direct tire pressure monitoring systems required in new light vehicles by the 2008 model year; see the National Highway Traffic Safety Administration rulemaking, Federal Register, Sept. 7, 2005, https://www.federalregister.gov/documents/2005/09/07/05-17661/federal-motor-vehicle-safety-standards-tire-pressure-monitoring-systems

10. I. Rouf and others, Security and Privacy Vulnerabilities of In-Car Wireless Networks: A Tire Pressure Monitoring System Case Study, USENIX Security Symposium (2010), finding a 32-bit static identifier readable from a passing vehicle at roughly 40 meters, https://www.usenix.org/legacy/events/sec10/tech/full_papers/Rouf.pdf. For the 2026 demonstration tracking vehicles over several weeks with low-cost software-defined-radio receivers, see the IMDEA Networks Institute study as reported, https://www.theautopian.com/the-signals-from-your-cars-tire-pressure-sensors-could-be-used-to-monitor-when-you-leave-home-report/

11. On Wi-Fi MAC address randomization in current mobile operating systems, see https://www.cablelabs.com/blog/mac-address-randomization-how-user-privacy-impacts-wi-fi-and-internet-service-providers

12. On defeating randomization, see Jeremy Martin and others, A Study of MAC Address Randomization in Mobile Devices and When It Fails, Proceedings on Privacy Enhancing Technologies 2017, issue 4, which defeated the technique on roughly 96 percent of the Android phones tested, https://petsymposium.org/popets/2017/popets-2017-0054.php. On the Bluetooth Low Energy weakness in Apple’s Continuity protocol, see Handoff All Your Privacy, https://arxiv.org/pdf/1904.10600

13. Hadi Givehchian and others, Evaluating Physical-Layer BLE Location Tracking Attacks on Mobile Devices, IEEE Symposium on Security and Privacy (2022), which fingerprinted devices in public spaces and tracked a volunteer to a residence using equipment costing under 200 dollars, while also documenting practical limits such as sensitivity to temperature; the same group published a mitigation in 2024, https://cseweb.ucsd.edu/~schulman/docs/oakland22-bletracking.pdf

14. SignalTrace product sheet (2025), Tracked Device Types, listing radio-frequency identification tags including key cards and pet microchips; see also the product page, https://www.leonardocompany-us.com/lpr/elsag-signaltrace

15. On passive low-frequency RFID and pet microchips, which carry no power source and transmit only when energized by a scanner held within a few centimeters, see https://enviroliteracy.org/can-nfc-read-pet-microchip/

16. Leonardo, ELSAG SignalTrace materials, https://www.leonardocompany-us.com/lpr/elsag-signaltrace

17. Leonardo US Cyber and Security Solutions press release, May 30, 2024, quoting general manager Jason Laquatra, https://www.businesswire.com/news/home/20240530567391/en/Leonardo-US-Cyber-and-Security-Solutions-receives-a-patent-for-their-new-electronic-signature-tracking-system-for-law-enforcement

18. On the company declining to comment beyond its marketing, see Carscoops, June 2026, https://www.carscoops.com/2026/06/signaltrace-plate-reader-tracking/

19. Carpenter v. United States, 585 U.S. 296 (2018).

20. United States v. Jones, 565 U.S. 400 (2012) (Alito, J., concurring in the judgment); Leaders of a Beautiful Struggle v. Baltimore Police Department, 2 F.4th 330 (4th Cir. 2021) (en banc).

21. Schmidt v. City of Norfolk (E.D. Va. Jan. 27, 2026), granting summary judgment to the city; see https://valawyersweekly.com/2026/02/16/norfolk-license-plate-reader-fourth-amendment-dismissal/

22. On the prevailing result and the reversal of the contrary Virginia trial-court ruling in Commonwealth v. Bell, see the Boston Bar Association survey, Eyes on the Road: AI, Privacy, and Automated License Plate Readers, https://bostonbar.org/journal/eyes-on-the-road-ai-privacy-and-automated-license-plate-readers/

23. Schmidt v. City of Norfolk (E.D. Va. Jan. 27, 2026), as reported, https://courthousenews.com/judge-holds-norfolks-license-plate-reader-use-constitutional/

24. Wiretap Act, 18 U.S.C. § 2510 and following; the definition of contents appears at 18 U.S.C. § 2510(8), and the exemption for radio communications readily accessible to the general public at 18 U.S.C. § 2511(2)(g)(i) and § 2510(16), https://www.law.cornell.edu/uscode/text/18/2510

25. Leonardo materials describing the capture of device frequencies emitted into the air, https://www.leonardocompany-us.com/lpr/elsag-signaltrace

26. Pen register and trap and trace statute, 18 U.S.C. §§ 3121 to 3127 (definition at § 3127(3); court order required at § 3121), https://www.law.cornell.edu/uscode/text/18/3127; Stored Communications Act, 18 U.S.C. § 2701 and following.

27. New York State Senate Bill S9890 (2025-2026), sponsor’s justification stating that there is no existing state framework restricting the use, transfer, storage, or sharing of plate-reader data, https://www.nysenate.gov/legislation/bills/2025/S9890

28. Chapter 632 of the Laws of 2025, amending Executive Law § 840 to require a minimum-standards policy and defining a plate-reader system as one that converts images of license plates into computer-readable data; see Senate Bill S963 (2025-2026), https://www.nysenate.gov/legislation/bills/2025/S963

29.New York State Senate Bill S9890 (2025-2026), which would impose use restrictions, an exclusionary rule, and penalties addressed to captured plate data, https://www.nysenate.gov/legislation/bills/2025/S9890

The first of those statements identifies a technical reality that the subsequent sections of this article document in detail: the target of CSI-based sensing does not need to own or operate any WiFi device, because the attacker supplies the transmitting hardware. The qualifying statement, that Bent had never personally deployed the technique, is worth noting precisely because the question is not whether any particular individual has done it. The question is whether the law has anything meaningful to say when someone can reconstruct a person’s body position, breathing rate, and heart rate from the other side of a wall, using hardware that costs $30, produces no detectable emission, and captures no image that any existing statute was written to address.

The answer, examined against the current state of federal law and constitutional doctrine, is almost nothing.

What WiFi DensePose Actually Does

In December 2022, researchers Jiaqi Geng, Dong Huang, and Fernando De la Torre at Carnegie Mellon University’s Human Sensing Laboratory completed a paper titled “DensePose From WiFi,” subsequently posted to arXiv as a preprint and submitted as a master’s thesis to the CMU Robotics Institute (arXiv:2301.00250).² The paper described a deep neural network that maps the phase and amplitude of WiFi signals to UV coordinates across 24 regions of the human body, reconstructing full-body human pose in real time, through walls, without a camera.

The mechanism is a property of WiFi signals called Channel State Information, or CSI. CSI is not the data transmitted over a network. It is a measurement of how the signal traveled from transmitter to receiver, capturing amplitude and phase across dozens of individual frequency channels called subcarriers. A human body moving through a room disturbs those signal paths in ways that are consistent, measurable, and recoverable. A neural network trained on those disturbances can determine where a person is, how they are positioned, and whether their chest is rising and falling.

In early March 2026, a developer known as Reuven Cohen released an open-source implementation of the technique under the name RuView, also known as WiFi DensePose, on GitHub under an MIT license.³ The repository reached the top of GitHub’s trending lists within days.

Figure 1. CSI-based through-wall human sensing at two capability levels. Left: the attacker deploys a WiFi transmitter and ESP32 receivers in an adjacent unit. Center: the CMU lab demonstrated a 24-region dense body surface map using research hardware (arXiv preprint, not peer-reviewed). Right: commodity ESP32 hardware, validated in peer-reviewed literature, produces a 17-keypoint skeleton with vital sign monitoring and has demonstrated through-wall activity recognition at 18.5 meters. The comparison table distinguishes what each capability level has actually demonstrated.

What It Actually Requires, and Why That Matters

Popular coverage has repeatedly misreported a critical technical distinction: a standard home router cannot perform this sensing. The GitHub repository states it directly. Full CSI-based pose estimation, vital sign monitoring, and through-wall detection require hardware that exposes per-subcarrier amplitude and phase data. Standard consumer WiFi equipment does not. The repository’s own documentation notes that consumer laptops “can only provide RSSI-based presence detection, which is significantly less capable.”³

What does work is an ESP32-S3 microcontroller, a commodity chip available from consumer electronics retailers for approximately $8 to $30. A functional sensing mesh requires three to six nodes plus a transmitting router; a documented starter configuration costs approximately $54 total. Alternatively, an Intel 5300 network interface card ($15 used, with modified firmware) or a Qualcomm Atheros AR9580 ($20 used, with kernel patches) can perform the same function.⁴

Peer-reviewed literature places the effective sensing range for commodity CSI hardware at approximately six to eight meters within a single room.⁵ Through-wall performance degrades with distance, but peer-reviewed research using ESP32-S3 hardware combined with a directional biquad antenna has demonstrated activity recognition across 18.5 meters spanning five rooms separated by 25-centimeter brick walls.⁶ A standard residential wall between adjacent units falls well within that range.

The Yagi question that prompted Bent’s response has a precise answer. A Yagi is a directional antenna that improves signal gain in one direction; it is relevant to transmitting or receiving radio signals, not to defending against sensing. CSI-based detection does not intercept data packets. It reads the physics of how radio waves scatter off a human body. WPA3 encryption is equally irrelevant: CSI operates at the physical layer of the radio signal, below the layer where any encryption is applied.⁷ What Bent identified correctly, even as an assumption, is that the attacker introduces their own transmitting hardware into the signal environment. The target does not need to own or operate any WiFi device. They need only to be present inside the signal path.

Figure 2. CSI operates at the physical layer (Layer 1) of the OSI model. Encryption protocols such as WPA3 protect payload data at Layer 2 and above. The amplitude and phase patterns of radio waves, which CSI reads, are unaffected by encryption.

One factual qualification is required before proceeding. The specific RuView/WiFi DensePose implementation has faced substantive criticism. Developers on Hacker News and GitHub identified code generating random CSI signals using np.random.rand() rather than actual hardware input, and alleged misrepresentation of cited research. Separate review flagged overnight inflation of the repository’s star count. The repository’s own audit log confirms that no real-world CSI dataset is bundled; the system runs on a synthetic reference signal.⁸ The underlying Carnegie Mellon research is a different matter. The CMU paper is an arXiv preprint and master’s thesis, not a peer-reviewed publication, but CSI-based human sensing as a field has been validated across hundreds of academic studies, deployed commercially at scale, and formally standardized by the IEEE under 802.11bf, published in September 2025.⁹ A two-year evaluation involving 280 edge devices across 16 real-world scenarios, processing more than four million motion samples, achieved 92.61% accuracy in diverse homes, and the resulting system now operates on over 100 million connected devices worldwide.⁹ Whether this specific GitHub implementation performs as advertised is a question about one developer’s repository. Whether the underlying capability exists is not. It does.

Two Separate Legal Problems

WiFi DensePose presents two distinct legal questions requiring separate analysis. The first is what happens when a government actor deploys it. The second is what happens when anyone else does. The answers are different, and both are incomplete.

Figure 3. The two-track legal analysis and the gaps within each track. For government actors, Kyllo and Carpenter provide constitutional grounds for requiring a warrant, but the “general public use” qualifier is unresolved and no statute specifically covers CSI sensing. For private actors, no federal statute prohibits it.

Track One: Government Actors

The Fourth Amendment stands, in the Supreme Court’s own formulation, most firmly at the threshold of the home. In Silverman v. United States, the Court declared that “[a]t the very core of the Fourth Amendment stands the right of a man to retreat into his own home and there be free from unreasonable governmental intrusion.”¹⁰ That principle has not weakened. Every major Fourth Amendment case addressing new surveillance technology in the decades since has returned to it.

The case that maps most directly onto WiFi DensePose is Kyllo v. United States, 533 U.S. 27 (2001). Federal agents used a thermal imaging device aimed at Danny Lee Kyllo’s home from a public street to identify heat patterns consistent with indoor marijuana cultivation. The Supreme Court, in a 5–4 opinion by Justice Antonin Scalia, held the surveillance was a Fourth Amendment search. The operative holding: where the government uses a device that is not in general public use to explore details of the home that would previously have been unknowable without physical intrusion, the surveillance is a “search” and is presumptively unreasonable without a warrant.¹¹

The majority opinion was written deliberately to reach forward. The Court declined to limit the rule to thermal imaging and warned that restricting it would “leave the homeowner at the mercy of advancing technology, including imaging technology that could discern all human activity in the home.”¹¹ Justice Scalia was describing a hypothetical risk. That hypothetical is now a $54 hardware kit running open-source software.

WiFi DensePose does not detect heat radiating through an exterior wall. It reconstructs body position, breathing rate, and heart rate of individuals inside the home, information that is by any defensible measure more intimate than the thermal distribution Kyllo addressed. Applied by a court willing to take the long view the majority invited, the Kyllo framework would require a warrant before government deployment of this technology against a private residence.

The open legal question is Kyllo’s “not in general public use” qualifier. The ESP32-S3 is a chip sold at consumer electronics retailers for as little as $8. Whether hardware at that price point, paired with a software stack available on GitHub, satisfies that threshold is a question no court has yet resolved. It bears noting that the Kyllo majority did not define “general public use” with precision, and the phrase has generated significant scholarly debate in the quarter-century since. A thermal imager that attaches to a smartphone is now available for approximately $200. Whether the availability of a technology in principle, as opposed to its widespread actual deployment for surveillance purposes, is the relevant metric is itself an unresolved doctrinal question.

A second line of authority bears on continuous monitoring. In Carpenter v. United States, 585 U.S. 296 (2018), the Supreme Court held that warrantless government acquisition of 127 days of historical cell-site location data constituted a Fourth Amendment search. Chief Justice Roberts’s majority opinion held that “an individual maintains a legitimate expectation of privacy in the record of his physical movements as captured through CSLI.”¹² That holding was tied to cell-site location information specifically, but the reasoning extends naturally to CSI-based sensing, which captures a qualitatively different and more intimate category of data: not where a person traveled over a period of months, but how their body is positioned, whether they are breathing, and what their heart rate is, inside a private home, in real time. Carpenter’s logic, applied to that record, resolves in one direction.

The statutory gap remains even under this favorable constitutional reading. Title III of the Omnibus Crime Control and Safe Streets Act of 1968, codified at 18 U.S.C. § 2510 et seq., and the Foreign Intelligence Surveillance Act, 50 U.S.C. § 1801 et seq., were both designed around the interception of communications content. CSI-based sensing does not intercept communications. It reads radio wave physics. No court has ruled on whether existing statutory authority covers it.¹⁴

Track Two: Private Actors

The Fourth Amendment constrains only government action. It says nothing about what private actors may do.

There is currently no federal statute in the United States that specifically prohibits WiFi-based covert human sensing.¹³ The Federal Communications Commission’s statutory mandate centers on spectrum efficiency and interference prevention, not on restricting the information that can be extracted from an authorized transmission. The Electronic Communications Privacy Act of 1986 and Title III govern the interception of the contents of wire, oral, and electronic communications. CSI-based sensing does not intercept any communication. It reads the physical behavior of radio waves as they propagate through space and interact with a human body, operating below the layer that federal wiretapping law was written to address.¹⁴

State law provides no consistent alternative. Some states have voyeurism or surveillance device statutes that courts might stretch to cover this conduct. None were written with through-wall CSI body mapping in mind, and most have not addressed it at all.

Any private actor who deploys an ESP32-S3 mesh and uses it to monitor body position, sleep patterns, and vital signs through a wall is operating in a legal space that existing law has not closed. The European Union’s General Data Protection Regulation classifies WiFi tracking identifiers as personal data, but CSI-based body pose extraction, which produces a real-time skeletal map of a specific individual inside a specific dwelling, has not been specifically addressed by regulators on either side of the Atlantic.¹⁵

The Question the Law Has Not Answered

WiFi DensePose is not a camera. It captures no image. It produces no recording a person would recognize as surveillance footage. What it produces is a skeletal data structure, 17 body keypoints updated in real time, and a continuous stream of physiological readings. That output is invisible in the legal sense that matters most: no existing surveillance statute was written to reach it.

Traditional surveillance law was designed around evidence that maps onto statutory categories legislators anticipated. A camera produces an image. A listening device produces audio. A tracking device produces location data. CSI-based sensing produces none of them. The sensing apparatus produces no detectable emission of its own, though the hardware nodes that enable it must be physically deployed in the signal environment. The data it extracts is not something the subject ever chose to expose. It is recovered from the involuntary physics of a human body occupying a room, inside the space the constitutional order has always treated as the most protected.

In Kyllo, Justice Scalia wrote that the Court must “take the long view.”¹¹ In 2001, the technology before the Court could determine whether someone was running high-intensity grow lights. The technology commercially available in 2026 can identify which room a person occupies, whether they are standing or supine, whether they are breathing at a normal rate, and what their heart rate is, through a wall, without their knowledge, for $54 in hardware.

The gap between those two facts is not a technical gap. It is a legislative one. And it is, at present, open.

Endnotes

1. Sam Bent (@DoingFedTime), posts on X, March 8, 2026. The first post (“Not having wifi does”) was published at 7:27 PM; the second (“The attacker brings the router…”) at 10:08 PM. Available at

2. Jiaqi Geng, Dong Huang, and Fernando De la Torre, Dense Human Pose Estimation From WiFi, CMU Robotics Institute Technical Report CMU-RI-TR-22-59 (August 2022); arXiv preprint arXiv:2301.00250 (submitted December 31, 2022). Available at https://arxiv.org/abs/2301.00250. This paper has not been published in a peer-reviewed journal or conference proceeding. It is a master’s thesis and arXiv preprint. It has been cited in subsequent peer-reviewed literature and forms the conceptual basis for the open-source implementation discussed in this article.

3. RuView GitHub repository (formerly WiFi DensePose), developer Reuven Cohen (ruvnet), MIT License. Available at https://github.com/ruvnet/wifi-densepose. Docker images and Rust crates were published on March 1, 2026; the repository reached GitHub’s trending lists by approximately March 5, 2026. See AIToolly, “RuView: WiFi DensePose for Real-time Pose, Vitals, Presence,” March 5, 2026. The repository states: “CSI-capable hardware required.” The repository’s documentation claims deployment via Docker “in approximately 30 seconds.”

4. On hardware costs: The RuView repository documents a starter kit of three to six ESP32-S3 nodes for approximately $54 total. Alternative CSI-capable hardware includes the Intel 5300 NIC ($15 used, with modified firmware) and the Qualcomm Atheros AR9580 ($20 used, with kernel patches). See Halperin et al., “Tool Release: Gathering 802.11n Traces with Channel State Information,” ACM SIGCOMM Computer Communication Review 41, no. 1 (2011): 53; Xie et al., “Commodity Wi-Fi-Based Wireless Sensing Advancements over the Past Five Years,” Sensors 24, no. 22 (2024): 7195, Table 11.

5. On sensing range: Xie et al., supra note 4, available at https://pmc.ncbi.nlm.nih.gov/articles/PMC11597943/ (“The existing sensing range is usually just 6–8 m within a single room”).

6. Julian Strohmayer and Martin Kampel, “WiFi CSI-Based Long-Range Through-Wall Human Activity Recognition with the ESP32,” in Computer Vision Systems: ICVS 2023, LNCS vol. 14253 (Springer, 2023), 41–50, https://doi.org/10.1007/978-3-031-44137-0_4. ESP32-S3 with 2.4 GHz biquad antenna; 18.5 meters across five rooms separated by 25 cm brick walls. Peer-reviewed conference proceedings.

7. CSI operates at the Physical Layer (Layer 1) of the OSI model. Encryption protocols such as WPA3 protect payload data at Layer 2 and above. See Xie et al., supra note 5, Section 2; Ma et al., “WiFi Sensing with Channel State Information: A Survey,” ACM Computing Surveys 52, no. 3 (2019): Article 46.

8. byteiota.com, “WiFi DensePose Hits GitHub #2: Real or AI-Generated Hype?” (March 2026), available at https://byteiota.com/wifi-densepose-hits-github-2-real-or-ai-generated-hype/. The repository’s own audit log (WITNESS-LOG-028) confirms: “Real-world CSI dataset bundled: No. Only synthetic reference signal (seed=42).”

9. IEEE Std 802.11bf™-2025, published September 26, 2025. See https://standards.ieee.org/ieee/802.11bf/11574/. On large-scale deployment: “Experience Paper: Scaling WiFi Sensing to Millions of Commodity Devices,” arXiv:2506.04322 (June 2025): two-year evaluation, 280 edge devices, 16 scenarios, 4+ million motion samples, 92.61% accuracy. System now deployed on 100+ million devices. See also NIST, “IEEE 802.11bf: Enabling the Widespread Adoption of Wi-Fi Sensing,” https://www.nist.gov/publications/ieee-80211bf-enabling-widespread-adoption-wi-fi-sensing.

10. Silverman v. United States, 365 U.S. 505, 511 (1961).

11. Kyllo v. United States, 533 U.S. 27 (2001). Opinion by Scalia, J. Operative holding at 533 U.S. at 40. “Long view” language at 533 U.S. at 35–36. Available at https://supreme.justia.com/cases/federal/us/533/27/.

12. Carpenter v. United States, 585 U.S. 296 (2018). Opinion by Roberts, C.J. “[A]n individual maintains a legitimate expectation of privacy in the record of his physical movements as captured through CSLI.” Slip op. at 11. The holding was tied to CSLI. Its extension to CSI-based sensing is the article’s analytical argument, not a direct application of the holding.

13. No federal statute enacted as of March 2026 explicitly addresses or prohibits covert CSI-based human sensing. The FCC’s authority under 47 U.S.C. § 301 et seq. centers on spectrum management, not information extraction. See 47 C.F.R. Part 15.

14. Title III, 18 U.S.C. § 2510 et seq. “Intercept” defined at § 2510(4) as acquisition of “contents” of communications. CSI sensing does not acquire contents; it reads physical propagation characteristics. Whether courts would extend “contents” to cover physical-layer body sensing is unresolved. See Justice Manual § 9-7.000, https://www.justice.gov/jm/jm-9-7000-electronic-surveillance.

15. GDPR (Regulation 2016/679) Art. 4(1), Recital 30 (WiFi identifiers as personal data). Whether CSI-derived biometric data falls under Article 9 special categories is unresolved.