Technology / The Privacy Desk
A VPN tunnel is not a cloak
A VPN is the most misunderstood product on the internet. It hides real things — and leaves other things in plain view that most people assume are covered. Here is exactly who can still see you, what happened when “no-logs” promises met courtrooms, and how the law changed underneath all of it.
Picture your internet traffic as postcards. Everyone who handles a postcard can read it. A VPN puts your postcards inside one envelope addressed to the VPN company, which opens it and mails the contents onward. Your mail carrier — your internet provider — now sees only that you send a lot of envelopes to one address. But the VPN company inherits the privileged position your carrier just lost: your real IP plus every destination and its timing, while HTTPS still seals the contents of properly encrypted pages. That is most of what a VPN is. The rest of this article is what that sentence does not cover.
Who sees what: the honest grid
Five observers watch your browsing. A VPN blinds exactly one and a half of them.
The second row is the one the ads skip. The VPN company inherits the privileged position your provider lost: your originating IP plus destination and traffic metadata for every connection — attached, usually, to a payment method with your name on it — while HTTPS still protects the contents of properly encrypted pages. Even Mozilla, which sells a VPN, states plainly that a VPN will not stop cookies, fingerprinting, or logged-in accounts from recognizing you. And your ISP still sees the tunnel itself: a real million-user ISP identified more than 85% of OpenVPN connections in live traffic.
The employer row deserves one plain sentence: on a company-managed laptop or phone, monitoring software sits below any VPN — it can capture the screen, the keystrokes, and the browser itself, and corporate firewalls with an installed company certificate can decrypt HTTPS outright. No consumer VPN changes that. (On an iPhone, Settings → General → VPN & Device Management shows whether such a profile is installed.)
A VPN is a trust decision, not an invisibility decision. You are choosing to believe a VPN company’s promises over your internet provider’s. Sometimes that is a good trade — the next sections are about checking the promises.
The two-session problem
Connect. Look at something. Disconnect. Reconnect an hour later. Can anyone tell both sessions were you? Three observers — and none of them necessarily forget you.
On observer one, the proof is a court record. In 2017, the FBI asked PureVPN — a “no-logs” provider — about a Massachusetts cyberstalking suspect. PureVPN’s records showed the same customer connecting from two originating IPs: his home and his workplace — and investigators described the same VPN exit IP touching his real Gmail, the burner account sending the threats, and an everyday pet-sitting account within minutes of each other. No browsing logs were needed; timestamps did it. Ryan Lin was sentenced to more than 17 years. The counter-model proves the rule: Mullvad issues random numbered accounts, takes no email, and accepts cash in envelopes precisely so it cannot link sessions to a person.
On observer three, the research is stark. DeepCorr, a deep-learning correlation attack presented at one of security’s top conferences, matched flows through an anonymity network with about 96% accuracy using roughly 900 packets — a few seconds of browsing — and that was against three-hop Tor. A one-hop VPN is the easier target. Website-fingerprinting studies hit ~98% accuracy guessing which site you visited from encrypted traffic patterns alone in lab conditions.
Turn the kill switch on — it is the setting that failed Sharp. Assume the provider can connect everything you do under one account. And know that for a motivated, well-positioned observer, timing patterns are a signature you cannot turn off.
Cookies do not care about your IP
The tracking industry stopped caring where you connect from years ago. It recognizes the browser itself.
A cookie is a name tag your browser volunteers to every site that set one. Change your IP a hundred times; the name tag stays pinned. And even with cookies blocked, your browser’s configuration — fonts, screen size, time zone, graphics quirks — forms a fingerprint. In the EFF’s landmark study of real browsers, more than 8 in 10 carried a fingerprint unique enough to identify them with no cookies at all.
Then there is the mode people trust most. Incognito windows delete local history when they close — and nothing else. Google spent four years litigating what its own servers collected from Incognito users, and settled by agreeing to delete or de-identify billions of records of private-browsing data, rewrite Incognito’s disclosures to say Google still collects data, and block third-party cookies in Incognito for five years. Class members received $0.
Privacy from websites is a browser problem, not an IP problem. Firefox isolates cookies per-site by default; Mullvad Browser (built with the Tor Project) makes your fingerprint generic; separate browser profiles keep your logged-in life away from your private one. A VPN complements these. It replaces none of them.
The iPhone section: Apple does it differently
iPhones have a private-browsing tool most owners already pay for — and a VPN quirk Apple has never fixed.
Start with the quirk, because it is the kind of thing this article exists for. In 2020, Proton disclosed that on iOS, connections opened before the VPN turns on keep running outside the tunnel — including Apple’s own services. Researcher Michael Horowitz confirmed it was still leaking in 2022; follow-up testing found Apple services bypassing VPNs into iOS 16; and as of his August 2025 update, he reports the behavior remains. Apple has called aspects of it expected behavior and offers developers a mitigation API. If your threat model is serious, the defensible summary is: an iPhone VPN app cannot promise that everything goes through the tunnel.
Private Relay’s split design means no single company holds the complete picture — architecturally stronger for Safari browsing than a one-hop VPN, and useless for anything outside it.
And the App Store itself is part of the threat model. The Tech Transparency Project found roughly 1 in 5 of the top 100 free VPN apps in the US App Store had obscured Chinese ownership — several linked to a US-sanctioned firm — while academic testing of free Android VPN apps found decades of consistent results:
On an iPhone: Private Relay for everyday Safari privacy, a reputable paid VPN when you need all-app coverage — knowing the iOS leak caveat — and never a free VPN from the charts. Check VPN & Device Management for profiles you did not install.
The courtroom record: when “no logs” met subpoenas
The marketing says nobody keeps logs. The court record keeps score — and it opens with a VPN that was the spy.
Two nuances the table cannot hold. First, jurisdictions can force the issue going forward: in 2021, Proton — under a binding Swiss order — began logging one account’s IP prospectively, leading to a French activist’s arrest. A no-logs policy is not immunity from a future court order. Second, breaches happen upstream: NordVPN’s 2018 incident was one rented server compromised through the datacenter’s management interface — forward secrecy meant no past traffic could be decrypted, but it took a researcher’s tweet in 2019 to surface it.
Can governments just decrypt the tunnel? The documented answer is: they did, when the crypto was weak. Snowden-era NSA documents describe an industrial VPN-exploitation pipeline — PPTP was broken outright, and researchers later showed a single weak 1024-bit prime would let a nation-state passively decrypt about 66% of IPsec VPN servers of that era. Modern, correctly configured WireGuard and OpenVPN have no known practical break; WireGuard’s handshake carries machine-checked proofs. The experts’ refrain: the math holds — so attackers go after keys, endpoints, metadata, and humans instead.
Pick providers whose no-logs claims survived contact with police — the table names four. Prefer WireGuard or modern OpenVPN. And treat coffee-shop Wi-Fi as the actual battleground: that is where TunnelVision-class attacks live, and where a VPN — ironically — earns its keep even while these bugs exist.
Can they find your address?
Not from your VPN. From your habits — that is a different story, and it runs through the dark web’s favorite spreadsheet.
First, the reassuring mechanics. An IP address geolocates to a city at best — commercial databases are right about the city only 50–75% of the time, with errors of 25–50 km — and it resolves to your internet provider’s equipment, not your door. Turning an IP into a street address requires a legal demand to the ISP. A website operator watching a VPN visit has none of that; they would need to identify the provider, compel it (often across borders), hope it kept logs, and then compel your ISP — four steps, each needing court authority.
Now the unreassuring mechanics. People do get found — through a chain that ignores the VPN entirely:
The darkest irony in the research: free VPNs feed this exact machine. SuperVPN’s 2023 leak exposed 360 million records pairing email addresses with users’ real IPs and locations — the precise join a stranger needs to connect “anonymous” activity to a person. The tool bought for privacy became the breach.
Can you scrub yourself out? Consumer Reports tested seven paid data-removal services against the people-search sites for four months. The results argue for doing it yourself:
Your address leaks through your email, your phone number, and forty old accounts — not your IP. Unique emails per site, a masked phone number, and an annual people-search opt-out sweep protect your address better than any VPN ever will.
The law moved: why 2025–26 changed everything
VPNs went from nerd tool to civic flashpoint in eighteen months — and lawmakers noticed.
The backdrop is American: in 2017, Congress killed FCC rules — finalized but never yet in effect — that would have required your consent before your internet provider used or sold browsing data, and barred similar rules from returning. Since then, only general consumer-protection law constrains what ISPs do with your history — the original mass-market reason to buy a VPN. What changed recently is age verification. 27 states (as of August 2026) require adult sites to verify visitors’ ages; the Supreme Court upheld Texas’s law 6–3 in June 2025; and Pornhub has blocked itself in roughly two dozen states in protest.
Then the UK ran the natural experiment. When the Online Safety Act’s age checks switched on July 25, 2025, VPN apps seized the UK download charts — Proton reported signups up 1,400% within hours, tracked UK VPN traffic ran roughly +1,300% to +2,000% over baseline for days, and five VPNs filled the top-10 free apps, including free ones researchers flag as dangerous. Ofcom’s position, as of mid-2026: VPN use by adults is legal and no ban is contemplated — but sites must not encourage children to circumvent checks.
And the crosshairs are real, if so far empty. Wisconsin’s age-verification bill originally required adult sites to block known VPN traffic — a US first — until backlash stripped the provision in February 2026. The UK House of Lords voted to ban VPN provision to under-18s in January 2026; the Commons rejected it in March. A federal bill that would force sites to treat VPN addresses as unverified remains stalled in committee as of publication — and Utah’s 2026 amendments point at the next frontier, judging location by physical presence rather than IP, so a VPN’s address no longer legally moves you.
Nothing here makes a VPN illegal for you in the US or UK. What the trend means: VPNs are now load-bearing civic infrastructure, governments are studying them, and the free-app sewers fill fastest exactly when a law sends millions of new users shopping. Choose before the rush.
The actually-private checklist
In order of effort. Each step closes a hole the previous ones cannot.
And to be fair to the much-advertised product: a VPN can earn its subscription on an untrusted network, against ISP data collection, and wherever your IP is the thing being judged — though with HTTPS now near-universal, even public Wi-Fi is less naked than the ads suggest. It is a good tool. It was just never a cloak.
How we measured this
Technical claims are cited to peer-reviewed research (USENIX Security, ACM CCS, PETS), CVE disclosures, and vendor documentation; legal cases to Department of Justice releases, court records, and first-party provider disclosures; legislation to statutes, the Supreme Court’s opinion, regulator statements, and primary trackers. Statistics that conflict across surveys (VPN adoption rates) use one labeled series. Where the research record has a gap — such as the absence of any documented end-to-end case of VPN activity being matched to identity via leaked databases — the article says so explicitly rather than dramatizing the mechanism.
This article describes documented capabilities and risks for a general audience. It is not legal advice, not an endorsement of any provider, and not a guide to evading any law. Product behaviors and laws described were verified as of August 19, 2026, and both change quickly.
Common questions
Can police track you through a VPN?
Does a VPN hide my browsing from my internet provider?
If I disconnect and reconnect to the same VPN, can the two sessions be linked?
Can someone find my home address from my IP or VPN?
Sources & references
Technical claims cite peer-reviewed papers and CVE disclosures; legal cases cite DOJ releases, court records, and first-party provider statements; legislation cites primary documents and regulator pages. Laws and product behaviors change quickly — details verified August 19, 2026.
- EFF Surveillance Self-Defense. What a VPN does and does not protect; the trust-shift to the provider; how Tor differs. ↗
- Mozilla. A VPN vendor’s own admission: VPNs do not stop cookies, fingerprinting, or logged-in account tracking. ↗
- DeepCorr (ACM CCS 2018). Deep-learning traffic correlation: ~96% flow-matching accuracy with ~900 packets, against Tor — a one-hop VPN is the easier case. ↗
- USENIX Security 2022. “OpenVPN Is Open to VPN Fingerprinting” — a real ISP identified >85% of OpenVPN flows; best-paper award. ↗
- Leviathan Security. TunnelVision (CVE-2024-3661): rogue DHCP routes traffic outside any routing-based VPN; Android unaffected; unfixable at the protocol level. ↗
- TunnelCrack (USENIX 2023). LocalNet/ServerIP attacks; every tested iOS VPN vulnerable at publication; vendor patches tracked. ↗
- Citizen Lab / PETS 2024. Port Shadow: attackers sharing your VPN server can intercept and de-anonymize peers. ↗
- Proton (2020) & Michael Horowitz (2022–25). The iOS VPN leak: pre-existing connections persist outside the tunnel; reported unfixed as of August 2025. ↗
- Apple. iCloud Private Relay technical overview: the two-hop design, and its Safari/DNS/insecure-HTTP scope. ↗
- Tech Transparency Project. Roughly 1 in 5 top free VPNs in the US App Store had hidden Chinese ownership; several tied to a sanctioned firm. ↗
- CSIRO / UNSW / Berkeley. Analysis of 283 Android VPN apps: 38% malware flags, 84% IPv6 leaks, 66% DNS leaks, 75% tracking libraries. ↗
- EFF Panopticlick. 83.6% of 470,161 tested browsers carried a unique fingerprint — no cookies required. ↗
- Brown v. Google settlement. Google agreed to delete or de-identify billions of Incognito-era records and rewrite its disclosures; $0 to class members. ↗
- DOJ. The PureVPN case: “no-logs” provider’s records linked two originating IPs to one customer; 17+ year sentence. ↗
- DOJ. HideMyAss logs, produced under UK court order, convicted the LulzSec Sony hacker. ↗
- TorrentFreak. IPVanish’s “zero logs” connection records, produced to Homeland Security in 2016, revealed via court documents. ↗
- Mullvad. April 2023: Swedish police arrived with a search warrant and left with nothing — no customer data existed to seize. ↗
- ExpressVPN. Turkey seized its server in the Karlov assassination investigation; the server held no logs. PIA (2016, 2018) and OVPN (2020) passed equivalent court tests. ↗
- Comparitech. UFO VPN and six sibling “no-log” free apps exposed ~1.2TB of activity logs, including plaintext passwords. ↗
- vpnMentor. SuperVPN’s 2023 exposure: 360 million records pairing emails with users’ real IPs and locations. ↗
- NordVPN. The 2018 single-server breach via a datacenter management interface; forward secrecy prevented retroactive decryption. ↗
- Logjam / weakdh.org (ACM CCS 2015). One weak 1024-bit prime would enable passive decryption of ~66% of IPsec VPN servers — the likely basis of Snowden-era NSA capability; PPTP was broken outright in 2012. ↗
- WireGuard. Machine-checked formal verification of the modern VPN handshake — the crypto that has no known practical break. ↗
- DOJ SDNY. The Ubiquiti case: a momentary VPN drop with no kill switch exposed Nickolas Sharp’s real IP; six-year sentence. ↗
- Proton. The 2021 Swiss order compelling prospective IP logging of one account — why “no logs by default” is not immunity from future orders. ↗
- National Public Data breach (2024). ~2.9B rows; roughly 272M SSNs with names, phones, and address histories — the dark web’s join table. ↗
- Consumer Reports (2024). Paid data-removal services removed 35% of found records over four months; manual opt-outs hit 70%. ↗
- Supreme Court. Free Speech Coalition v. Paxton (June 27, 2025, 6–3): Texas’s adult-site age-verification law upheld under intermediate scrutiny. ↗
- Free Speech Coalition tracker. 27 states with adult-site age-verification laws as of August 2026, with effective dates. ↗
- UK surge reporting. Proton signups +1,400% within hours of the Online Safety Act’s July 25, 2025 age checks; tracked UK VPN traffic up ~1,300–2,000%; Ofcom: VPN use legal, no ban contemplated. ↗
- EFF. Wisconsin’s first-in-nation VPN-blocking mandate, stripped from the bill in February 2026 after backlash; the UK Lords’ under-18 VPN ban was rejected by the Commons in March 2026. ↗
- CERT-In (India). The 2022 directive requiring 5-year VPN customer logs — major providers removed physical Indian servers rather than comply. ↗
- S.J.Res. 34 (2017). The repeal of FCC rules that would have required consumer consent for ISP use or sale of sensitive browsing data, and that bars substantially similar FCC rules from returning — the origin of mass-market VPN adoption. ↗
- FTC (2021). Staff report: major ISPs combine browsing, app, and location data into ad segments, including sensitive categories. ↗
- Brennan Center. FISA Section 702 lapsed in June 2026; existing FISC certifications keep collection authorized into March 2027. The CFPB withdrew its data-broker rule May 2025. ↗
- Security.org. Annual US consumer survey: 32% of US adults currently use a VPN (2025 series; survey methodologies vary widely). ↗
- MaxMind. IP geolocation accuracy: city-level correctness of 50–75% with errors of 25–50 km — an IP is not an address. ↗
- Unsealed court records (via TechCrunch). Facebook’s Onavo VPN and “Project Ghostbusters”: intercepting Snapchat analytics traffic at Zuckerberg’s urging — the VPN as the spy. ↗
- FTC. Avast sold “re-identifiable” browsing data from its privacy software to 100+ third parties; $16.5 million in redress. ↗
- US Treasury / Europol (2026). “First VPN Service”: 12 years of no-logs marketing while police secretly captured its traffic for 4½ years; 33 servers seized in May, sanctions in July. ↗
- DOJ. The 911 S5 botnet spread via free VPN apps: 19M+ device IPs rented to criminals; $5.9B in confirmed fraud. ↗
- DOJ. Genesis Market: 80M credentials bundled with victims’ cookies and device fingerprints from 1.5M computers — browser identity for sale. ↗
- Mysk (Aug 4, 2026). Three WebKit features — DNS prefetch, WebAuthn, WebTransport — leak real IPs past iCloud Private Relay; Apple fix planned for fall 2026. ↗
- FTC (2026). Kochava banned from selling sensitive phone-location data that could trace individuals to homes, clinics, and churches — location leaks from GPS, not IPs. ↗
- California Privacy Protection Agency. DROP: one deletion request reaching 600+ registered data brokers; first Delete Act enforcement announced August 11, 2026. ↗
- Consumer Reports. 12 of 16 tested VPNs overstated protections; lawmakers separately asked the FTC to probe misleading claims like “military-grade encryption.” ↗