Are Free VPNs Safe in 2026?

Most free VPNs are not safe to use. Free VPN apps have been documented logging and selling user data at scale, shipping broken or absent encryption, and, in one documented case, reselling users’ bandwidth as a commercial proxy network. A small number of free tiers from audited, subscription-funded providers are safe, but they are the exception rather than the rule.

The distinction that matters is not whether a VPN costs money. It is how the provider recovers the cost of running its infrastructure. VPN infrastructure is expensive: servers across multiple countries, bandwidth at scale, engineering, and security audits all carry ongoing costs. A provider with no subscription revenue must recover those costs somewhere, and the mechanism it chooses determines the risk. Without a VPN, your traffic is distributed: your ISP sees every destination you connect to, each service you use sees only its own slice, and no single commercial party holds the whole record with an incentive to sell it. A VPN concentrates the ISP’s view into one place and hands it to a company operating under fewer regulatory constraints. If the entity holding it monetises what it sees, the VPN has not reduced your exposure; it has redirected it.

Free tier, free trial, or free app: which one are you looking at?

Four distinct products get called a free VPN, and the risks apply primarily to one of them.

A permanent free tier of a subscription provider (freemium). The provider sells a paid product and offers a capped free version on the same infrastructure. The free tier is funded by paying subscribers, not by its own users. This is the only category in which a free VPN has no financial incentive to exploit user data. Proton VPN Free, Windscribe Free, PrivadoVPN Free, hide.me Free and TunnelBear Free all operate on this model.

A time-limited free trial of a paid provider. This is the full paid product on a timer, typically seven to thirty days, usually requiring payment details to activate. The risk profile is the paid provider’s, not the standalone free app market’s. Using a free trial of an audited paid VPN provider differs from installing a standalone free app.

A money-back guarantee window. Most reputable paid VPN providers offer a thirty-day money-back guarantee. Conditions vary: some exclude short or promotional plans, some cap data usage during the refund window, and purchases made through the Apple App Store or Google Play are governed by the platform’s refund policy rather than the provider’s. Read the specific terms before relying on this. Where it applies cleanly, the result is an audited VPN at zero net cost for a defined period; an option most users searching “are free VPNs safe” are unaware of.

A standalone free app with no paid product. No subscription tier, no disclosed revenue model, no apparent way to sustain the service other than its users. This is the category where the risks are concentrated. The research on free VPN security sampled the market as a whole, mostly the most-installed free apps on Google Play, which includes the free tiers of subscription providers alongside standalone apps. None reported failure rates separately for standalone free apps and freemium free tiers. The case for freemium tiers being different rests on their funding model and audit evidence, not on that research having excluded them.

One clarification on terminology: a free VPN is not the same as a free web proxy. A proxy reroutes browser traffic at the application layer and does not encrypt the payload. A VPN creates an encrypted tunnel at the network layer covering all device traffic.

Why free changes the risk: how free VPNs make money

Running a VPN server network costs money regardless of what a provider charges its users. Servers in multiple countries, bandwidth at scale, software engineering, and periodic independent security audits are recurring expenses. A provider that charges nothing has three options for recovering those costs: a paying subscriber base that subsidises the free product, advertising and in-app tracking, or monetising user data directly. The funding model is the primary risk signal for any free VPN, and it is almost always determinable before installation.

Three distinct funding models operate in the free VPN market. The freemium model funds a capped free tier through revenue from paid subscribers, who use the same infrastructure. No incentive exists to monetise free users’ data, because doing so would damage the paid product. The advertising and tracking model embeds third-party SDK libraries in the app that collect behavioural data and transmit it to advertisers, regardless of what the tunnel itself carries. The data monetisation model logs browsing activity, connection metadata, or device identifiers and sells them to data brokers or advertising networks; the tunnel may work correctly while the provider’s backend collects and sells what passes through it.

Connecting to any VPN shifts trust from your ISP to the VPN provider. Your ISP can no longer see your traffic destinations; the VPN provider now can. This shift improves your privacy only if the provider’s practices are better than your ISP’s. For a provider operating on the advertising or data-monetisation model, the shift transfers surveillance from a regulated telecommunications operator to a less-accountable commercial entity with a direct financial incentive to exploit what it observes. The full implications of this trust shift are covered in the cyber threats guide. The same logic applies to paid VPNs: payment removes the data-monetisation incentive; it does not verify that a provider has no records to produce. A paid VPN with no published independent audit makes the same unverifiable claims as a free VPN with a privacy policy statement.

Four types of dangerous free VPN

The category determines the specific threat, what the user experiences, and whether the problem is detectable. A given app can fall into more than one category simultaneously.

Malware and technical failure

Some free VPN apps install malware, spyware, or adware alongside or instead of any VPN function. Others provide genuine tunnelling but fail at the cryptographic or routing level, transmitting traffic in a form that is readable to third parties.

Configuration file vulnerability (current, highest documented severity). A peer-reviewed study published at NDSS 2026 (MVPNalyzer, Wang et al.) found five free Android VPN apps that fetched their VPN configuration file over cleartext HTTP. An attacker on the same local network can intercept and rewrite that configuration file before it reaches the app, redirecting all of the user’s traffic to a server the attacker controls. The researchers built and confirmed the attack. The researchers later named the five apps to The Hacker News; a July 2026 retest reported there found VPN Pro and Hexa VPN no longer fetching the file in cleartext, while BambooVPN, Free VPN (package ID com.appoxide.freevpn) and 101 VPN remained vulnerable.

Malware flags. A peer-reviewed analysis of free Android VPN apps by researchers at CSIRO’s Data61, UNSW, ICSI, and UC Berkeley (Ikram et al., ACM IMC 2016, data from September 2015) found that 38% of apps triggered at least one antivirus flag. The researchers applied their own threshold of five or more independent detections before treating an app as likely malicious; only 4% of apps cleared that bar. Reporting the 38% figure without the 4% qualifier overstates what the researchers themselves concluded by nearly tenfold. A 2024 industry test of the 100 most-installed free Android VPN apps (Top10VPN, June 2024) found 19 triggered warnings on VirusTotal. Those are scanner flags, not confirmed malware; most came from a single engine, and the most common flag is possibly a false positive linked to Google’s AdMob library, though the report notes many apps containing AdMob did not trigger it.

One example: sFly Network Booster, a traffic-optimisation app using the Android VPN permission, carried a Spyware signature and was one of ten apps at or above the authors’ five-detection threshold (flagged by 10 VirusTotal engines, tying for sixth place in the corpus ranking). It requested READ_SMS and SEND_SMS permissions, which the researchers noted could be used to read users’ messages and, potentially, send texts to premium-rate numbers. The app had approximately 1,000 installs and was delisted from Google Play by August 2016.

No-encryption tunnelling. The CSIRO-led 2016 study found that 18% of tested apps implemented tunnelling with no encryption at all, functioning as basic TCP or HTTP forwarders that transmitted all traffic in plaintext.

Both the malware-flag and no-encryption figures come from Android apps tested against a September 2015 dataset, predating Play Store Data Safety labels, Android’s runtime permission model changes, and near-universal HTTPS adoption. These figures reflect a market that has changed; the failure mode persists, but the prevalence numbers are not current.

Hard-coded cryptographic credentials. A 2025 peer-reviewed study (Mixon-Baca, Knockel, and Crandall, “Hidden Links: Analyzing Secret Families of VPN Apps,” FOCI 2025) identified three families of free VPN apps with combined Google Play downloads exceeding 700 million. In two of those families, the researchers found Shadowsocks passwords hard-coded directly into the APKs. Extracting the password requires nothing more than the publicly downloadable APK. Anyone who can then observe the traffic anywhere between the user and the VPN server, whether on the same local network or at an ISP or transit network, can decrypt it. Several apps used RC4-MD5 for Shadowsocks, a cipher suite deprecated since 2020.

Blind on-path attacks. The same FOCI 2025 study found that all three app families were also vulnerable to a separate failure: a network adversary on the same connection can infer a user’s active connections through blind on-path attacks despite the tunnel appearing active. It requires network adjacency and does not involve decryption, distinct from the hard-coded password issue.

A Zimperium zLabs analysis of approximately 800 free VPN apps published in October 2025 found three apps still shipping OpenSSL versions vulnerable to Heartbleed, a critical cryptographic flaw first disclosed in 2014.

Traffic leaks. Free VPNs fail at basic tunnel integrity at higher rates than paid alternatives. DNS leaks, where DNS queries travel outside the tunnel, were found in 24 of 281 tested apps by MVPNalyzer: in most cases they reached the local network’s resolver, and in six apps they reached public resolvers such as Google’s and Cloudflare’s. The CSIRO-led 2016 study found 66% of tested Android apps failed to route DNS through the tunnel (data from 2015). The 66% figure reflects a much older sample alongside MVPNalyzer’s current data; the two are not directly comparable, but both confirm the pattern. WebRTC leaks follow the same pattern through a different browser-level mechanism.

Risk: Device compromise, credential theft, financial data exposure, enrollment in a botnet, or decryption of tunnelled traffic by any party who extracts the hard-coded key from the publicly available APK and can observe traffic anywhere between the user and the VPN server.

Data logging and selling

A data-logging VPN provides genuine tunnelling. The user’s IP address changes, and a leak test returns clean results. The privacy failure happens in the provider’s backend: browsing activity, DNS queries, connection timestamps, IP addresses, and device identifiers are recorded and sold to advertisers, analytics companies, or data brokers. You have replaced ISP surveillance with provider surveillance; the provider has a more direct financial incentive to use what it collects.

The failure is not detectable from the user’s side. No VPN leak test, no network monitor, and no browser extension can reveal that a provider is logging and selling connection data from its own servers. The verification methods that exist are independent audits of server infrastructure, documented legal cases where the provider either had or did not have records to produce, and infrastructure design choices that make retention technically impossible. All three are covered in the guide to what a no-log VPN is and how to verify the claim.

Risk at scale. A 2024 industry analysis of the 100 most-installed free Android VPN apps (Top10VPN, June 2024) found that 71 were observed actually sending personal data to third parties during testing, including advertisers, social platforms, and data brokers such as Kochava. A separate October 2025 analysis by Zimperium zLabs of approximately 800 free VPN apps across iOS and Android published no figure for third-party data transmission, but found compliance and transport-security failures at scale: 25% of iOS apps lacked a valid Apple privacy manifest and over 6% requested private entitlements, which Zimperium treats as a red flag; it did not confirm that any were granted.

The white-label shared infrastructure problem. Many free VPN apps that appear independent share a single backend: the same servers, the same database, and the same developer entity under different brand names. A data sale or breach at one exposes all of them simultaneously, and users have no way to identify the shared infrastructure from the app store listing. The most extensively documented case, involving seven Hong Kong-based providers sharing a single exposed database of over one billion records, is covered in the no-log VPN guide.

In February 2021, a breach of the free Android VPN apps SuperVPN and GeckoVPN exposed over 20 million records, including email addresses, login timestamps, device make and model, IMSI numbers, and device serial numbers. A small number of FlashVPN records appeared in the same file; HIBP noted that this suggests the three brands may share a common platform; that is an inference from data co-location, not a confirmed finding. Separate reporting on the hacker-forum sale of the data also listed ChatVPN. SuperVPN had approximately 100 million installs.

Risk: Browsing history sold to data brokers; behavioural profiling; records available to law enforcement if a compelled-disclosure request is made to the provider.

In-app tracking and traffic manipulation

A third category monetises user behaviour through mechanisms operating at the app layer rather than the network layer. The goal is the same as data logging, to sell information about users to advertisers, but the mechanism bypasses the tunnel entirely.

Current form: SDK-layer tracking. Providers embed third-party tracking SDK libraries (software modules that collect device and behavioural data on behalf of advertisers) directly in the VPN application. These collect device identifiers, usage patterns, and behavioural signals and transmit them through the app’s own telemetry channels. Because this data exits through the app rather than through the tunnel, it is invisible to any leak test or network monitor. The Top10VPN June 2024 analysis found 84 of 100 apps contained tracking SDKs from marketing or social-media platforms; in 80, that code included at least one privacy-risking function alongside the permission needed to run it. Approximately 20 contained code from ByteDance or Yandex with the corresponding device permission to transmit data to those firms.

Historical form: traffic injection. The original documented version of this type involved injecting advertising or tracking code directly into unencrypted HTTP traffic. The CSIRO-led 2016 study identified two apps doing this. Hotspot Shield, operated at the time by AnchorFree Inc., redirected user traffic at e-commerce sites including Alibaba, eBay, Target, Best Buy, and Macy’s through affiliate partner redirect chains to Conversant Media and Viglink. WiFi Protector injected JavaScript into users’ pages for advertising and tracking; its developers confirmed this to the researchers when contacted.

In August 2017, the Center for Democracy and Technology filed a complaint with the FTC asking it to investigate AnchorFree for unfair and deceptive trade practices. For the JavaScript injection via iframes, more than five third-party tracking libraries, and e-commerce redirects, the filing cited the Ikram et al. study. Its own new evidence was a static analysis using Carnegie Mellon University’s Mobile App Compliance System, which found undisclosed sharing of data including Wi-Fi network names, MAC addresses, and device IMEI numbers with third-party advertising networks. CDT set these findings against AnchorFree’s public claims of “Anonymous Browsing” and CEO David Gorodyansky’s statement that “we never log or store user data.” The FTC does not publicly confirm investigations, and no enforcement action against AnchorFree has been documented. AnchorFree denied the allegations at the time. On the injection and redirect findings, the CDT filing is not independent corroboration: it relies on the Ikram et al. study. The peer-reviewed measurement is the primary evidence; the petition itself remains unresolved.

The injection-into-cleartext-HTTP mechanism is now largely obsolete. With HTTPS near-universal, unencrypted HTTP traffic represents a small and shrinking fraction of web requests.

Risk: Behavioural data, device fingerprints, and usage patterns transmitted to third-party advertising platforms through channels the tunnel does not govern; in the historical injection model, modification of the content delivered to the user.

Bandwidth reselling

A fourth category monetises the free user’s network connection rather than their data. The provider enrolls free users’ devices as exit nodes in a commercial proxy network. Paying customers purchase access to those residential IP addresses and route their own traffic through them. You are providing infrastructure for activity you did not authorise and cannot see.

Documented case: Hola VPN. Hola VPN routed free users’ connections through peer devices and sold access to those residential IP addresses through Luminati, its commercial proxy arm. In 2017, UK private-equity firm EMK Capital bought a majority stake in Luminati, with Hola’s founders retaining a significant stake. Luminati rebranded as Bright Data in March 2021 and now operates as a separate company from Hola. Hola’s own FAQ states that the service works by sharing users’ contributed resources for business use by Bright Data: specifically, idle bandwidth from the free Windows app and from Android versions distributed outside Google Play. Its browser extensions, iOS app, and Google Play Android app do not participate. Hola Premium is the opt-out.

In late May 2015, a Luminati user operating under the handle Bui, who told 8chan founder Fredrick Brennan he had signed up for a free trial of the service, used it to attack 8chan. Thousands of legitimate-looking POST requests hit 8chan’s post.php endpoint in thirty seconds, a hundredfold spike that crashed the site’s PHP-FPM process. Because the requests arrived from ordinary residential and university IP addresses belonging to Hola’s free users, none of 8chan’s spam filters flagged them. Hola founder Ofer Vilenski confirmed to The Register that his network was the infrastructure the attacker used, and said the company adjusted its Luminati screening process afterwards. This is an application-layer denial-of-service whose effectiveness depended on the cleanliness of the source IPs, not on bandwidth volume. The attacker reached those IP addresses through Luminati’s commercial service, on a free trial.

Following Brennan’s public disclosure of the incident, Hola amended its FAQ to state explicitly that free users act as exit nodes for paying Luminati customers. The arrangement had been present in the terms of service before this amendment but was not prominently disclosed.

What the user experiences: Elevated data usage, slower connection speeds, and traffic originating from their IP address that they did not generate.

Risk: Traffic purchased through the user’s IP address that is used for activity the user did not authorise, including large-scale scraping, ad fraud, or denial-of-service attacks, some of which is unlawful and all of which is attributable to the account holder for that IP.

Browser extensions: a different product with a different failure mode

A free VPN browser extension is not a smaller version of a free VPN app. It is architecturally different, with different risks, and none of the major studies of free VPN security measured it.

A browser extension VPN functions as an application-layer proxy, not a system-level encrypted tunnel. It routes browser traffic through the provider’s servers while leaving all other device traffic (system services, other applications, operating system telemetry) on the regular connection. If you install a free VPN browser extension to “protect your traffic,” you have protected only one of several traffic streams on your device.

The permission model for browser extensions compounds this. A VPN extension typically requests permission to read and change all your data on every website you visit. This is the same permission an extension needs to inject content, exfiltrate credentials, or log every page the user visits. It is granted at installation with a single click, with no indication of what the extension does with it. An extension from a provider in the data-monetisation or SDK-tracking category has both the access and the commercial incentive to use that permission against the user’s interest.

MVPNalyzer (NDSS 2026), Ikram et al. (ACM IMC 2016), Mixon-Baca et al. (FOCI 2025), Top10VPN (June 2024), and Zimperium zLabs (October 2025) all analysed VPN apps, not browser extensions. Their findings cannot be assumed to apply to extensions. This is a limitation of the evidence base, not a reason to assume extensions are safer; the funding-model logic and the absence of independent audits apply to free VPN extensions with equal force.

A fifth risk: covert ownership and jurisdiction

Beyond what a provider does with user data, a fifth risk concerns who the provider is and what legal obligations its true ownership creates, regardless of its stated privacy policy.

A VPN provider is legally subject to the laws of its jurisdiction of incorporation. In jurisdictions with broad national intelligence laws, providers can be compelled to log and surrender user data under non-disclosure requirements. The intelligence-sharing alliances that matter most for VPN jurisdiction are the Five, Nine, and Fourteen Eyes arrangements. Covert ownership, where the entity visible in the app store is a shell company fronting a parent in a compulsion-heavy jurisdiction, extends that legal exposure to users who cannot identify it from the app listing.

A months-long investigation published by Proton VPN on 27 August 2026, drawing on download data from AppTweak and tracker-detection data from Exodus Privacy (an open-source Android app scanner), examined 390 VPN applications available through the Apple App Store and Google Play in the US. It found 64 owned by Chinese companies, with 31 concealing ownership behind shell companies registered in Singapore, Hong Kong, and the UK. Disclosure: Proton AG is a commercial VPN provider, and the report concludes with a pitch for Proton VPN’s own service. Proton names its sources for download estimates (AppTweak) and tracker detection (Exodus Privacy, Android only) but does not describe how it established ownership of the 64 apps. Its ownership counts have not been independently replicated; the pattern they describe is consistent with the one corporate chain that TTP’s filing analysis and the FOCI 2025 paper trace in detail.

A Tech Transparency Project investigation published in April 2025, examining AppMagic’s ranking of the 100 most-downloaded free VPN apps in the U.S. App Store during 2024, found that 20 were owned by companies or individuals in mainland China or Hong Kong without clearly disclosing it. Those apps had been downloaded more than 70 million times from U.S. app stores, per AppMagic estimates. The Tech Transparency Project is a research initiative of the Campaign for Accountability.

TTP’s filing analysis traces Turbo VPN, VPN Proxy Master, Thunder VPN, Snap VPN, and Signal Secure VPN through a network of Singapore and Cayman Islands holding companies (Innovative Connecting Pte. Ltd., Lemon Seed Technology Ltd., Lemon Clove Pte. Ltd., and Autumn Breeze Pte. Ltd.) to Qihoo 360 (360 Security Technology Inc.), which disclosed acquiring all three app companies in its 2019 annual report to the Shanghai Stock Exchange.

Qihoo 360 was added to the U.S. Commerce Department’s Entity List in June 2020 over concerns about military end-use procurement, restricting U.S. exports to the company without a licence. Separately, Qihoo 360 was designated a “Chinese military company” on the Defense Department’s Section 1260H list in October 2022; it remains on that list as of the June 2026 update. The Section 1260H designation is not a sanction; it triggers DoD procurement prohibitions and compliance consequences.

Qihoo 360 reported selling the app companies to unnamed external parties in September 2020. TTP found March 2025 corporate filings suggesting a continuing connection. TTP describes the apps as “recently owned by a sanctioned firm” and says it found evidence suggesting Qihoo 360 remains connected to them. The “sanction” TTP refers to is the Commerce Department’s Entity List action; strictly, that is an export-control listing rather than a sanctions designation. Current ownership is not publicly established.

TTP’s filing analysis and the FOCI 2025 cryptographic analysis converge on the same corporate network from two independent methodological directions: corporate filings and shared server credentials. They corroborate one documented chain, not two separate discoveries. The Proton investigation is a broader prevalence survey across a larger sample; it does not trace this particular chain. The three sets of figures come from different sample frames and cannot be added together.

What the user experiences: Nothing visible. An app covertly owned by a Chinese-incorporated entity may perform well technically while being subject to Chinese national intelligence law, which can compel cooperation with state agencies under non-disclosure requirements.

Risk: State intelligence access to traffic metadata, DNS queries, browsing history, or session data via legal compulsion to a parent entity the user cannot identify from the app store page.

What the research actually shows

The evidence on free VPN security comes from five bodies of research. Three are peer-reviewed academic papers or peer-reviewed workshop publications. Two are industry reports. They are not methodologically equivalent, and treating them as equivalent is the primary mechanism by which statistics in this area get distorted. All five studies also share a scope limitation: they measured VPN apps, not browser extensions. Their findings cannot be assumed to apply to the extension category.

MVPNalyzer, NDSS 2026 (peer-reviewed)

Wang, Ortwein, Sobrados, Stanley, Sharma, Anwar, and Ensafi. NSF-funded. Institutions: University of Michigan, University of New Mexico, and IIT Delhi. 281 free Android VPN apps from Google Play, dataset collected in November 2024, with results presented at NDSS in February 2026.

Findings: 29 of 281 apps leaked in at least one category. 24 specifically leaked DNS queries; those 24 apps account for approximately 360 million combined installs. 6 leaked full user traffic. 4 transmitted data through unencrypted tunnels. The configuration-file attack was built and confirmed. The researchers named the five apps to The Hacker News; a July 2026 retest reported there found three still vulnerable. This is the most current peer-reviewed research on free VPN security.

Ikram et al., ACM IMC 2016 (peer-reviewed)

Ikram, Vallina-Rodriguez, Seneviratne, Kaafar, and Paxson. Institutions: CSIRO’s Data61, UNSW, ICSI, and UC Berkeley. Data collected September 2015. 283 apps identified; 150 dynamically tested.

Key figures: of all 283 apps analysed, 38% triggered at least one AV flag; 4% cleared the authors’ own five-detection threshold. Of the 150 apps tested dynamically, 18% implemented tunnelling with no encryption; 16% deployed non-transparent proxies modifying HTTP traffic; 66% failed to route DNS through the tunnel. Two apps injected JavaScript into user traffic. This study predates Play Store Data Safety labels, Android’s runtime permission model changes, and near-universal HTTPS adoption. The underlying business model problem remains valid; the specific percentages reflect a different market.

Mixon-Baca, Knockel, and Crandall, FOCI 2025 (peer-reviewed workshop)

“Hidden Links: Analyzing Secret Families of VPN Apps.” Free and Open Communications on the Internet (FOCI) workshop. The 2025 edition appeared in the proceedings of the Privacy Enhancing Technologies Symposium (PETS). FOCI is a peer-reviewed academic workshop rather than a full conference.

Three families of free VPN apps identified, with combined Google Play downloads exceeding 700 million. In two of the three families, the researchers found Shadowsocks passwords hard-coded into the APKs; the third was only partially analysed. Anyone who extracts the key from the publicly downloadable APK and can observe the traffic between users and the VPN server can decrypt it. All three families were also vulnerable to blind on-path attacks. The Innovative Connecting / Autumn Breeze / Lemon Clove family, the one linked to Qihoo 360, is one of the three; the paper states it “has been previously scrutinized by two research efforts linking them to the People’s Liberation Army.”

Top10VPN, June 2024 (industry research)

Top10VPN.com is a commercial VPN review site that earns affiliate revenue from VPN recommendations published in the same report. One researcher. Android only. Sample: the 100 most-installed free VPN apps from Google Play, with combined installs of approximately 2.5 billion.

Figures: 88 of 100 apps leaked data outside the tunnel, most commonly DNS queries. 71 were observed actually sending personal data to third parties. 84 contained tracking SDKs from marketing or social-media platforms; in 80, that code included at least one privacy-risking function alongside the permission needed to run it. Approximately 20 contained code from ByteDance or Yandex with the matching device permission. 19 triggered malware flags on VirusTotal; these are scanner flags, not confirmed malware. It is not peer-reviewed research.

Zimperium zLabs, October 2025 (industry research)

Zimperium is a mobile security firm; its commercial interest is in the mobile security software market, not VPN recommendations specifically. Approximately 800 free VPN apps on iOS and Android: the only one of the five that covers both platforms.

Findings: approximately 1% bypass TLS certificate validation entirely. 25% of iOS apps lacked a valid Apple privacy manifest. Over 6% of iOS apps requested private entitlements, which Zimperium treats as a red flag; it did not confirm that any were granted. Three apps shipped OpenSSL versions vulnerable to Heartbleed.

What the five sources establish together

The peer-reviewed work establishes structural failures: broken or absent encryption, traffic manipulation, configuration vulnerabilities, and active data leakage at the tunnel level. The industry research documents data-sharing behaviour across larger app sets. Neither supports the undifferentiated claim common in published coverage: that most free VPNs are dangerous in the same way. They support bounded statements about distinct failure categories.

Safe free VPNs exist: the conditions that make them different

A free VPN differs when its free tier is subsidised by revenue from a paid product, covered by the same independently audited no-logs policy, and has no financial incentive to monetise free users. Five freemium tiers meet, to varying degrees, the criteria for a safe free option. They are not recommendations; the right choice depends on the user’s threat model and each provider’s current audit status.

Proton VPN Free

What it offers: Unlimited data, no ads, one simultaneous connection, servers in several countries. The kill switch, which cuts your internet connection if the VPN drops so traffic never travels unprotected, is available on the free tier in both its standard and permanent variants; Proton’s free plan page lists this as a core free-tier feature.

Why it differs: Proton AG generates revenue from paid Proton VPN subscriptions and other Proton products. The free tier is a loss-leader for the paid product. No incentive exists to sell or exploit free users’ data.

Audit: Securitum conducted an on-site no-logs audit at Proton AG’s Zurich offices from 20 to 22 May 2026. The scope explicitly covered selected free-tier and paid servers, configuration files, and change-control procedures. No mechanism creating or retaining user-identifiable activity records was found. This was the fifth consecutive annual engagement. A separate SOC 2 Type II attestation was completed in July 2025. The no-logs report is publicly downloadable without creating an account; Proton provides the SOC 2 report on request.

Audit scope note: The Securitum engagement is point-in-time and operator-assisted: Proton engineers demonstrated systems while Securitum directed verification. It is not a fully independent remote examination.

Jurisdiction: Switzerland (Proton AG, Plan-les-Ouates, Geneva). A proposed revision to Switzerland’s surveillance ordinance (VUPF) remains pending as of September 2026. If adopted, it would extend identification and retention obligations to derived service providers above a defined user threshold, a category that includes VPNs. The proposal drew broad opposition during consultation and remains unadopted. Proton has confirmed it began moving some infrastructure outside Switzerland ahead of any such change.

Disclosure: Proton AG also published that covert ownership investigation. Its broad finding is consistent with independent research; its ownership methodology has not been independently verified.

Limitations: One simultaneous connection; P2P (peer-to-peer file sharing, such as BitTorrent) not available on free servers (paid-plan feature); limited server selection; lower speeds on free servers.

Windscribe Free

What it offers: Up to 10GB per month, unlimited simultaneous connections, servers in roughly ten countries (the free location list changes periodically; check Windscribe’s current list), the R.O.B.E.R.T. ad and tracker-blocking feature with a reduced custom blocklist set on the free tier, and a Firewall (kill switch) on Windows, Android, and Linux; macOS and iOS users have an Always-On feature instead. The 10GB allowance requires a confirmed email address and continued subscription to Windscribe’s promotional emails. Without email confirmation, or after unsubscribing from marketing emails, the monthly allowance drops to 2GB. This condition is not prominently disclosed at sign-up.

Audit: No standalone independent no-logs audit has been published for Windscribe. The company has open-sourced its apps and completed a penetration test. A penetration test identifies security vulnerabilities; it does not verify whether connection records are created or retained on the servers. The two are not equivalent.

P2P: Blocked on free servers. P2P requires a paid Windscribe plan. Windscribe’s knowledge base states this plainly, and so does its own guide to P2P VPNs. Its server-list page promotes P2P use and says almost all locations support it, without noting that free servers are excluded. For current policy, the knowledge base is the authoritative source.

Jurisdiction: Canada, a founding Five Eyes member.

Limitations: Conditional and low data allowance; no no-logs audit; P2P blocked on free; Five Eyes jurisdiction.

PrivadoVPN Free

What it offers: 10GB per month, one simultaneous connection (apps can be installed on unlimited devices), 12 server locations. The kill switch is available on the free tier. P2P is permitted across all 12 free-tier server locations; PrivadoVPN does not document this policy explicitly, but independent hands-on reviews report P2P working on free servers. SOCKS5 proxy is restricted to paid plans.

Audit: No independent no-logs audit has been published in approximately six years of operation.

Jurisdiction: PrivadoVPN began relocating its legal home from Switzerland to Iceland in 2025 in response to the proposed Swiss surveillance revisions, and completed the move in 2026: Privado Networks ehf., Gardabaer (near Reykjavik), with revised terms of service effective 1 May 2026. A Swiss entity is retained for payment processing, which explains why the provider’s own no-log feature page still describes the service as Switzerland-based as of September 2026. Verify the current legal position against the privacy policy or terms of service rather than the features page. Iceland is outside all Eyes intelligence-sharing alliances and has no mandatory data retention framework applicable to VPN providers.

PrivadoVPN’s free tier states a no-logs policy. That policy has not been independently verified. Using PrivadoVPN Free means trusting the policy without third-party confirmation, which matters most when P2P is in use and your connection metadata is most exposed.

Limitations: No no-logs audit; one simultaneous connection; 10GB monthly cap; SOCKS5 proxy paid-only.

Hide.me Free

What it offers: A free tier with no email address or payment card required. Unlimited data on the free tier (confirmed first-party 8 September 2026; hide.me’s help documentation states the previous 10GB cap has been removed). 8 locations are available on the free tier. 1 simultaneous connection. The kill switch is available on the free tier, confirmed first-party: hide.me’s own published content states “all our VPN clients include this feature for both free and paid users.” On speed, hide.me’s pricing page lists “Restricted speeds” as a free-plan attribute, while hide.me’s own blog states it does not place any speed restrictions on its free VPN servers. Both are current first-party statements; hide.me offers no explanation for the discrepancy. Independent testing reports free-tier speeds below the paid tier; whether that is deliberate policy or server load is not established.

P2P is permitted on free-tier servers in the Netherlands, Germany, and select US locations; it is blocked on servers in Australia, Brazil, India, Mexico, South Korea, and other locations due to datacenter and regulatory requirements. The hide.me client displays P2P-compatible indicators next to each server.

Audit and transparency: Hide.me has commissioned two independent no-logs audits. The first was conducted in 2015 by DefenseCode Ltd., a Croatian application-security firm; hide.me states DefenseCode certified that it does not permanently store identifiable user activity or logs. The scope and method of that engagement were not published. At eleven years old, it is historical context rather than current evidence. Hide.me was, by industry and its own accounts, among the first VPN providers to commission an independent audit. The most recent was conducted by Securitum (the same firm that audits Proton VPN annually) between 19 March and 6 June 2024, with Securitum’s report published 7 June 2024. The audit was conducted remotely from Securitum’s offices in Poznan and Gdansk, Poland, based on a server environment snapshot provided by hide.me plus VPN configuration files, team interviews, and a review of the deployment process.

Securitum confirmed that all VPN servers are deployed from the same container image, making the snapshot analysis representative. Securitum concluded that hide.me’s declared no-logs policy was applied in practice and that no issues were found on VPN containers that could compromise user privacy. The audit also found scripts using iptables and the nDPI library on the container, designed to detect BitTorrent traffic and block it locally. They were inactive at the time of the audit due to their own script logic. Securitum’s conclusion was that automatic classification of this kind does not pose a threat to user privacy. The report does not connect these scripts to hide.me’s per-location P2P availability.

In late 2025, hide.me was re-accredited under the VPN Trust Initiative’s Trust Seal programme for 2026. The VTI is an industry-led consortium run by the i2Coalition, a trade body; its seal requires membership and a review against five principle areas (security, privacy, advertising, disclosure and transparency, and social responsibility), assessed by the i2Coalition itself and reaccredited annually. It is a self-regulatory industry accreditation rather than an independent technical examination of server infrastructure, and it does not verify a no-logs claim in the way the Securitum engagement does.

The Securitum audit refers to the system state as of 15 March 2024. As of September 2026, that reference state is approximately 30 months old, significantly beyond 18 months, the maximum age at which an audit should be treated as current. The audit remains materially stronger evidence than an unaudited policy statement, but the gap from Proton’s May 2026 engagement is substantial.

Hide.me also publishes an annual transparency report, every year since 2012. The 2025 report documents 9 law enforcement requests received and states: “We don’t store any data, so whilst we do respond to every request made, we have nothing to hand over.” Nine requests, zero user data produced: a real-world test of the no-logs policy under live law enforcement conditions.

Jurisdiction: eVenture Limited, Labuan, Malaysia. Malaysia is outside the 5/9/14 Eyes intelligence-sharing alliances.

Limitations: Two no-logs audits (2015, 2024) versus Proton’s five consecutive annual engagements; the 2024 audit is approximately 30 months old as of September 2026 and does not meet the 18-month currency standard; free-tier speeds are lower than paid in practice, and hide.me’s own pricing page and blog contradict each other on whether this is a deliberate restriction; P2P only on select servers; 8 locations versus the full paid network; 1 simultaneous connection.

TunnelBear Free

What it offers: 2GB per month for most users; users in countries with significant internet censorship may receive up to 10GB per month through TunnelBear’s anti-censorship programme. The kill switch (VigilantBear) is documented by TunnelBear as a feature of its Windows, macOS, iOS, and Android apps, with no plan restriction stated in the feature documentation; TunnelBear’s pricing page does not list it under either plan, so free-tier availability is not confirmed first-party. Current hands-on reviews report it working on the free tier. P2P permissions for the free tier are not documented by TunnelBear as of September 2026. As of 2026, free users cannot manually select a server; TunnelBear assigns the server automatically. Split tunnelling has moved to paid tiers. Simultaneous connection count is reported as unlimited in multiple sources, though TunnelBear’s plan comparison lists unlimited devices only under the paid plan.

Audits: TunnelBear has published eight consecutive annual security audits by Cure53, starting from 2017 (covering 2016). The most recent covered 2024 and was published in November 2025. No audit covering 2025 has been published as of September 2026.

Audit type distinction: These are security audits: white-box assessments of TunnelBear’s apps, backend infrastructure, and implementation. They identify security vulnerabilities and assess implementation quality. They are not no-logs audits. TunnelBear’s logging practices have not been independently verified as Proton’s and hide.me’s have.

Jurisdiction: Canada, a founding Five Eyes member. Canada differs from hide.me (Malaysia) and Proton (Switzerland) on this dimension.

Ownership: McAfee acquired TunnelBear in 2018. McAfee is a separate company from Gen Digital (the group that includes Norton, Avast, and Avira).

Limitations: Very low data allowance for most users; no manual server selection; split tunnelling paid-only; Five Eyes jurisdiction; eight consecutive security audits confirmed and no no-logs audit.

Comparison

The common condition shared by all five: subscription revenue from paid tiers funds the free product. The degree of independent verification differs. Proton has the strongest record: five consecutive annual no-logs audits with free-tier server scope confirmed, unlimited data, a kill switch on the free tier, and an audit currency within 18 months, noting that each Securitum engagement is point-in-time and operator-assisted rather than a fully independent remote examination.

Hide.me has one current no-logs audit (Securitum, 19 March–6 June 2024) and an earlier 2015 DefenseCode engagement of unpublished scope, unlimited data confirmed first-party, a kill switch confirmed first-party, an annual transparency report with zero user data produced in response to nine 2025 law enforcement requests, and P2P on select servers; the Securitum audit refers to the system state as of 15 March 2024, approximately 30 months before September 2026, and does not meet the 18-month currency standard. Windscribe and PrivadoVPN have no no-logs audit; TunnelBear has eight consecutive security audits and no no-logs audit. Treating all five as equivalent overstates the verification record of four of them.

Free VPN vs paid VPN: what the money actually changes

Payment removes one structural risk (the financial incentive to monetise user data) but does not remove all risk. A paid VPN provider with no published independent audit makes the same unverifiable no-logs claim as a free provider with a policy statement. The money buys a different incentive structure, not a guarantee.

The verification methods for any VPN’s no-log claim are the same regardless of price: independent audits with full server-infrastructure scope, documented legal cases where the provider produced nothing, and RAM-only server architecture. The full framework is in the no-log VPN guide; it applies to paid and free providers equally.

What “safe” means depends on your threat model

Whether a given free VPN is safe depends on what you are protecting against. “Safe” is not a binary property of a provider: the same provider is adequate for one use case and insufficient for another.

Casual use on public Wi-Fi (low-stakes). The threat is network-level monitoring or interception on a shared connection. A reputable freemium free tier handles this. Proton VPN Free is adequate and has no data cap. On capped tiers such as Windscribe, PrivadoVPN, or TunnelBear, the monthly allowance constrains extended use; the protection against passive network interception holds in all cases.

Ongoing privacy from ISP surveillance (moderate-stakes). A standalone free VPN with no clear business model is counterproductive here. You have replaced ISP surveillance, which operates within a telecommunications regulatory framework with defined disclosure requirements, with provider surveillance that may carry fewer constraints and a direct financial incentive to exploit the data collected. A freemium tier or a paid subscription with an audited no-logs policy is the minimum for this goal.

Sensitive communications: journalists, activists, source protection (high-stakes). A free VPN of any category is insufficient. A single commercial no-log VPN provides network-layer privacy against passive interception and ISP logging; it does not provide anonymity. The provider still sees traffic destinations, connection timing, and DNS queries, whether or not those are logged. For the distinction between privacy and anonymity, see the Tor vs. VPN guide. For evaluating any VPN against a high-risk threat model, see the full verification framework in the no-log VPN guide.

Is a free VPN safe for banking, torrenting, or streaming?

Online banking

HTTPS already provides end-to-end encryption for the content of banking sessions. The bank’s TLS certificate and the browser’s certificate verification cover the session content regardless of whether a VPN is running.

Two risks arise from a free VPN in a banking context. First, approximately 1% of free VPN apps bypass TLS certificate validation entirely (Zimperium zLabs, October 2025). This is a failure in the app’s own network code rather than a route into the browser’s banking session, which the bank’s certificate and the browser’s validation continue to protect. It indicates an app whose developers do not implement basic transport security correctly, a signal about everything else the app handles, including credentials entered into any in-app browser or webview the app controls. Second, a data-logging provider records which financial institutions the user connects to and when, even if it cannot read the content of the session.

Some banks’ fraud-detection systems flag known VPN IP ranges and may lock accounts or require additional verification steps.

Verdict: A free VPN with TLS validation failures poses risks to in-app browsing. A provider in the data-logging category exposes connection metadata but not HTTPS session content. Fraud-detection lockouts are a common practical consequence.

Torrenting

Safe torrenting with a VPN requires three things at once: a verified no-logs policy independently audited at the server level, a kill switch that prevents traffic from exiting the tunnel if the VPN connection drops, and P2P permission. Of the five free tiers, hide.me Free is the only one that appears to satisfy all three simultaneously, each condition carrying a caveat.

†The 2024 no-logs audit does not meet the 18-month currency standard. It remains materially stronger evidence than an unaudited policy statement, but it is not equivalent to Proton’s May 2026 engagement.

‡Kill switch confirmed first-party (hide.me, verified 8 September 2026).

§P2P available on Netherlands, Germany, and select US servers as of September 2026. Blocked on servers in Australia, Brazil, India, Mexico, South Korea, and other locations due to datacenter regulations. The hide.me client displays P2P-compatible indicators per server. Free-tier P2P is not published as a formal provider commitment; check the client’s server indicators before relying on it, and verify the policy has not changed since September 2026.

Proton VPN Free has a verified no-logs audit and a kill switch, but blocks P2P on free servers outright. This is enforced technically: the server rejects P2P traffic and the app surfaces an error.

Hide.me Free has a Securitum no-logs audit, a kill switch confirmed first-party, and permits P2P on select free-tier servers. The 2024 Securitum audit found scripts using iptables and the nDPI library on the audited container, designed to detect and block BitTorrent traffic locally but inactive at the time of the audit; Securitum concluded they do not threaten user privacy. It also publishes an annual transparency report: in 2025, nine law enforcement requests were received and no user data was produced, corroborating the no-logs claim under real conditions.

The caveats: the 2024 no-logs audit refers to the system state as of 15 March 2024, approximately 30 months before September 2026, and does not meet the 18-month currency standard; the audit was based on a container snapshot provided to Securitum rather than live on-site server access; P2P access depends on which server the user connects to (8 total free locations, P2P permitted on Netherlands, Germany, and select US nodes); free-tier P2P is not a formal documented provider policy and could change without announcement. None individually disqualifies it; together they set a lower bar.

PrivadoVPN Free has a kill switch and permits P2P on all 12 of its free servers, as reported in independent hands-on reviews. The provider does not document this as policy, which means it could change without notice. It has no independent no-logs audit. The claim that matters most when your IP address and traffic are visible to every peer in the swarm is the one nobody has checked.

Windscribe blocks P2P on free servers; P2P requires a paid plan. Its server-list page promotes P2P without noting that free servers are excluded; its knowledge base and its own P2P guide are explicit on the restriction.

Verdict: Hide.me Free is the only one of the five that appears to combine a verified no-logs audit, a kill switch, and P2P permission, subject to its audit-age and server-availability caveats. PrivadoVPN Free permits P2P with a kill switch but without a verified audit. For sustained P2P activity where the no-logs claim needs independent verification meeting the 18-month currency standard, a paid provider with an audited no-logs policy and a documented kill switch is the more robust choice.

Streaming

Streaming is a capability question more than a safety question. Streaming platforms maintain blocklists of known VPN IP addresses and update them continuously. Free VPN IP ranges, being heavily shared across many users, are among the first blocked. Effectiveness for accessing geo-restricted content is low and declining.

Data caps make sustained streaming impractical on most free tiers. Netflix’s help centre puts standard definition at up to 1GB an hour and HD at up to 3GB per hour. A 2GB monthly cap covers about two hours of standard-definition video; a 10GB cap covers a little over three hours of HD, for the entire month across every device connected to the account. Proton VPN Free and hide.me Free are exceptions: both offer unlimited data. Hide.me’s pricing page lists “Restricted speeds” as a free-plan attribute, while hide.me’s blog states no speed restrictions are applied; independent testing reports free-tier speeds below the paid tier regardless of which first-party source is correct. Proton VPN Free uses lower-speed free-tier servers than its paid tier. For both providers, the practical ceiling is speed rather than data volume.

The safety consideration is the provider’s logging practices.

Verdict: Free VPNs are ineffective for streaming due to IP blocklists and data caps. The relevant safety question is the provider’s logging practices, which apply regardless of what the user is doing online.

Free VPNs on work devices and in censored jurisdictions

Two groups face additional constraints: people installing a VPN on a device they do not own, and people for whom a free VPN is the only accessible option.

Employee-installed free VPNs and shadow IT

An employee who installs a free VPN on a corporate device routes all device traffic, including corporate application traffic, email, and internal system connections, through an unvetted third party. Corporate network monitoring, which the organisation uses for security incident detection, is bypassed for any traffic the VPN tunnels. The covert-ownership findings make this risk acute: an app whose ownership traces to a Chinese-incorporated entity, installed on a corporate device, creates a supply-chain exposure that goes beyond personal privacy.

The remedy is policy and managed deployment: an acceptable-use policy plus a managed VPN provider approved by IT. Per-app evaluation is not the right tool for an IT team that cannot audit individual staff installations.

When a free VPN is the only accessible option

In jurisdictions with significant internet censorship, a free VPN is frequently the only accessible tool. Paid VPN providers are often blocked by the same filtering infrastructure that restricts other content, and international payment methods may be unavailable. “Use a paid VPN” is not actionable advice for this audience.

Within the free category, what matters is provider selection: audited freemium tiers from providers with a structural reason to protect user data. TunnelBear offers users in censored countries up to 10GB per month through its anti-censorship programme, though free users cannot select servers manually, limiting its usefulness where specific locations are blocked. Proton VPN Free’s unlimited data removes the practical constraint that makes most free tiers unusable for sustained circumvention. The decisive variable in a censored network is not jurisdiction but whether the provider’s protocol survives the filtering in use locally; this varies by country and changes without notice. The Swiss jurisdiction Proton operates under is subject to the same pending surveillance ordinance (VUPF) that has not yet been adopted.

For users in censored jurisdictions who need stronger anonymity than a commercial no-log VPN provides, the Tor vs. VPN guide covers the relevant architecture and its trade-offs.

How to evaluate a free VPN before installing

Six checks tell you what you need to know before installing a free VPN. For paid VPNs, the full no-log verification framework is in the no-log VPN guide. Apply these in order; the first check resolves most cases.

1. Find the business model. Is there a paid tier generating enough revenue to sustain the free product? A standalone free app with no paid product, no disclosed revenue model, and no explanation of how the service is funded has no sustainable income source other than its users. The absence of a credible revenue model is the strongest signal that user data is the product.
2. Read the privacy policy for specific data fields. Field-level commitments (“we do not store originating IP addresses, DNS queries, or connection timestamps”) are checkable claims. Broad assurances (“we respect your privacy” or “we do not sell your data”) are not. A provider can collect connection metadata and never sell it; the relevant question is what is collected, not what is sold.
3. Check for an independent audit that is current (within 18 months) and covers the server infrastructure, not only the mobile app. An audit of the app alone says nothing about what the servers log. Know which type of audit it is: a no-logs audit verifies retention practices; a security audit verifies implementation quality. Infrastructure transparency, including RAM-only server architecture, is an additional signal within this check.
4. Check the permissions the app requests. A VPN needs network access. It does not need access to contacts, camera, microphone, or call log. Each permission beyond network access is either a data collection mechanism or evidence of poor engineering practice.
5. Check ownership, not just the brand name. Many free VPN apps share parent companies, server infrastructure, and databases under different brand names. Search the developer name shown in the app store, not just the app’s brand name, and cross-reference with public corporate registries. A consistent pattern of Singapore, Cayman Islands, or Hong Kong holding companies with no disclosed ultimate parent warrants investigation before installation.
6. Search for documented incidents. The provider’s name combined with “data breach,” “lawsuit,” or the names of law enforcement agencies. Documented breaches, court cases, and regulatory complaints are a matter of public record. A clean search result is not confirmation that the provider is safe; it is the absence of a disqualifying signal.

What to do if you have already used a free VPN

Uninstalling the app is not sufficient in every case, and some of what a free VPN collected cannot be recovered.

Remove the VPN configuration, not just the app. On iOS, deleting a VPN app from the App Store usually removes its VPN configuration with it; apps installed from the App Store use the system VPN framework, and iOS clears the configuration on uninstall. It is still worth confirming. Open Settings > General and look for VPN & Device Management (some iOS versions label it VPN, DNS & Device Management). Any VPN configurations still present appear under VPN; tap the ⓘ beside one and choose Delete VPN. If a configuration profile is listed on the same screen, open it and choose Remove Profile; this applies if the VPN was set up manually via a downloaded profile rather than from an app.

On Android, uninstalling the app normally removes its VPN configuration as well. To confirm, open Settings > Network & internet > VPN (see Google’s Android help), or search your device’s Settings for “VPN,” since the exact path differs across Samsung, Xiaomi, and Pixel devices. Tap the gear beside any remaining entry: if it opens the VPN app, that entry belongs to the app and will disappear when the app is uninstalled; if it shows an editable profile, that one was added manually and requires Forget to remove.

For a browser extension, remove the extension from the browser’s extension settings and separately revoke its site permissions, which can persist after the extension itself is removed.

Consider which credentials were exposed. If the app fell into the malware category, treat passwords for any account accessed while connected as potentially exposed; an app that can execute code on the device can capture input before it reaches the network, regardless of transport encryption. If it fell into the no-encryption category, HTTPS still protected credentials submitted to sites using it; what was exposed is the same metadata your ISP would have seen: which sites you visited and when, plus the contents of any remaining plain-HTTP requests. In either case, address email accounts first (email is a recovery mechanism for everything else), then financial accounts, then any other account where exposure would be consequential.

What cannot be undone. Data transmitted to advertising networks or data brokers during the period of use cannot be recalled. If a data-logging provider recorded browsing activity, that record exists in the provider’s systems and potentially with third parties it was sold to. The pre-install check matters more than any post-install response.

Legal exposure from bandwidth reselling. Traffic that originated from your IP address through a bandwidth-reselling provider is attributable to you in the first instance. If that traffic generates a copyright notice, a spam complaint, or a law enforcement enquiry, it arrives addressed to the account holder for that IP. The legal framework for VPN use, including the limits of IP attribution as evidence, is covered in the guide to whether VPNs are legal. The practical step: document that you were running a bandwidth-reselling VPN during the relevant period if a complaint arrives.

After switching providers, run a DNS leak test.

Frequently asked questions

Are free VPNs safe for everyday use?

It depends on which free VPN and what everyday use involves. A freemium free tier from a provider with an audited no-logs policy and a paying subscriber base is safe for routine tasks: browsing on public Wi-Fi, masking your IP from visited sites, preventing your ISP from logging your destinations. A standalone free VPN app with no paid product, no audit, and no disclosed revenue model is not safe for everyday use; its business model requires monetising what it observes.

Can a free VPN steal my data?

Yes, through several documented mechanisms. A provider in the data-logging category records and sells browsing history, connection metadata, and device identifiers. A provider embedding third-party tracking SDKs transmits behavioural data through the app’s own channels. A provider whose apps bundle malware can capture credentials, financial data, and device contents.

What is the safest free VPN?

Among free options, Proton VPN Free has the strongest verification record: five consecutive annual no-logs audits by Securitum, the most recent from May 2026 covering free-tier servers and meeting the 18-month currency standard. Hide.me Free also has a Securitum no-logs audit (June 2024, server infrastructure in scope) and adds P2P access on select servers, a confirmed kill switch, and no email required at sign-up; the 2024 audit is approximately 30 months old as of September 2026, which matters for threat models where verification currency is a criterion. Both rest on verified claims rather than policy statements alone.

Is a free VPN better than no VPN?

Not necessarily. A free VPN from a provider operating on the data-monetisation model shifts your traffic from your ISP to a provider with a financial incentive to exploit it and fewer regulatory constraints. In that case, using no VPN leaves your traffic with your ISP, which operates within a telecommunications regulatory framework. A free VPN is better than no VPN only if the provider’s practices are better than your ISP’s, which requires verification.

Do free VPNs sell your data?

Many do. Of the 100 most-installed free Android VPN apps tested in 2024, 71 were observed actively sending personal data to third parties during testing, including advertisers, social platforms, and data brokers. The user’s IP address changes; the data flow does not stop.

Are free VPNs safe on iPhone and Android?

The risks are the same on both platforms, though most large-scale research has focused on Android. A 2025 analysis of approximately 800 free VPN apps found that iOS apps are not exempt from these issues: a quarter lacked a valid privacy manifest, and over 6% requested private entitlements, which Zimperium treats as a red flag. The provider’s funding model, not the user’s platform, is the primary determinant of risk.

What free VPNs should I avoid?

A funding-model and audit check is more reliable than any named list. Specific apps change ownership, update their practices, or rebrand; a named avoidance list goes stale. The signals that indicate a free VPN to avoid: no paid tier and no disclosed revenue model; no published independent audit of server infrastructure; permission requests for contacts, camera, microphone, or call log; a developer entity that resolves to a shell company with no disclosed ultimate parent; or a documented breach or legal case where the provider produced records it claimed not to retain.

Do free VPNs slow down your internet?

Yes, to a degree that depends on the provider’s infrastructure and your chosen server. Every VPN adds latency through the extra routing hop to the VPN server and adds encryption overhead to each packet. Free VPNs introduce two additional factors: free-tier servers are shared across more users than paid servers, increasing congestion; and the nearest available free-tier server may be geographically further from the user than a paid server. The practical result is slower speeds and higher latency than a paid provider’s optimised infrastructure, though the difference matters most for latency-sensitive activities such as gaming and real-time calls.

Are paid VPNs automatically safe?

No. Payment removes the data-monetisation incentive that makes standalone free apps structurally unsafe. It does not verify that a provider has no records to produce, that its servers are correctly configured, or that its no-logs claims have been independently confirmed. A paid provider without a published independent audit makes claims no more verifiable than a free provider with a privacy policy. Apply the same criteria to a paid provider before subscribing.

Is a free VPN trial the same as a free VPN?

No. A free trial gives you the full paid provider’s infrastructure, policies, and audit record on a time limit, typically seven to thirty days. A money-back guarantee works similarly: full paid-product access during the window, with conditions that vary by plan and platform. The provider’s practices are those of a subscription business, not a provider that must monetise its free users.

Are free VPN browser extensions safe?

They carry the same structural risks as free VPN apps, plus a distinct failure mode. A browser extension VPN proxies only browser traffic; all other device traffic continues on the regular connection. The broad site-access permission most VPN extensions trigger, which lets them read and change data on every site you visit, gives the extension the same access it would need to log every page visited or inject content into any site. None of the five major studies of free VPN apps (MVPNalyzer, Ikram et al., Mixon-Baca et al., Top10VPN, Zimperium) measured browser extensions. Their findings apply to VPN apps; the extension category has not been studied at equivalent scale by peer-reviewed researchers. The absence of data is not evidence of safety.