How the Great Firewall of China Detects and Blocks VPNs
The Great Firewall of China is not a single wall, a static blocklist, or a switch that flips a site on or off. It is a distributed system of inspection points at China's international internet gateways that reads traffic patterns in real time, actively tests connections it finds suspicious, and adapts within hours of a new circumvention tool showing up. That last part is why no VPN, protocol, or provider has ever offered a permanent, guaranteed way through it. Anything that works today is working because it hasn't been caught yet, not because it's uncatchable.
In short: The Great Firewall blocks traffic through a layered combination of domain filtering, DNS manipulation, and a system that actively probes suspected VPN servers to confirm what they are before blocking them. Since 2021 it has also targeted any traffic that simply looks too random to be normal browsing, which is exactly why VPN protocols keep having to disguise themselves as ordinary web traffic instead of just encrypting it. No VPN protocol or provider is permanently reliable, the legal risk to an individual user is real but rarely enforced, and the bigger practical danger for most people is installing a "free" VPN app secretly owned by the kind of company the firewall is supposed to be protecting them from.
How Does the Great Firewall Actually Work?
The Great Firewall sits on-path at China's international gateways, meaning it mirrors passing traffic and can inject forged packets into it, but it cannot decrypt genuine encrypted traffic the way a true man-in-the-middle attack would. It relies instead on what it can see without decrypting anything: connection metadata, handshake structure, and increasingly, how random the traffic looks.
Domain and SNI filtering. When your browser connects to a site over HTTPS, the very first message (called the ClientHello) includes the domain name in plain text, in a field called SNI, before encryption is actually established. The Great Firewall reads this field and, if the domain is on its blocklist, forges a reset packet to sever the connection immediately. This filter alone covers hundreds of thousands of domains and is updated continuously rather than left static; one academic measurement tracking it over 20 months found the HTTP-layer filter growing by roughly 528,000 domain additions a month. In April 2024, the same technique was extended to QUIC, the newer connection protocol behind HTTP/3: because QUIC's initial handshake keys are derivable by any observer by design, the firewall can decrypt the opening packet, read the domain inside it, and block on that basis too. It's the first documented case of any country doing this to QUIC traffic at national scale, according to researchers at GFW Report, the academic group that has published most of the peer-reviewed research this section draws on.
DNS poisoning versus IP blocking. These are two separate mechanisms that get conflated. DNS poisoning is a race: when your device asks a DNS server "what's the address for this domain," an on-path device closer to you forges a fake answer and wins the race back to your device before the real answer arrives, so you never learn the correct address in the first place. A 20-month academic study called GFWatch catalogued over 1,700 IP addresses used in these forged responses. IP blocking is different: it lets your device resolve the correct address, then blocks traffic to that address directly. The firewall maintains largely separate blocklists for each method, and most of its blocking is short and reactive rather than a permanent ban, typically 120 to 180 seconds for a specific connection after something trips a filter.
The "fully encrypted traffic" rule. This is the development that shaped nearly every VPN protocol built in the last five years. In November 2021, the firewall started applying a different kind of test: instead of trying to positively identify VPN traffic, it checks whether a connection's first data packet looks like normal, structured protocols such as TLS or HTTP. If it doesn't, meaning the bytes look close to statistically random, which is exactly what strongly encrypted data looks like, the connection gets flagged and blocked. Genuine TLS traffic has a lot of structure to it (predictable header bytes, recognizable patterns) and doesn't trip this test. A VPN protocol whose handshake is just clean encrypted noise does. This single rule is why protocols like Shadowsocks and VMess, and even Tor's obfuscation layer, started failing in China after 2021, and why every serious circumvention tool built since has focused on looking like ordinary traffic rather than simply hiding what it says.
Why Some VPN Protocols Get Blocked and Others Don't
Every generation of VPN protocol built to get around the Great Firewall has eventually been caught by one specific, identifiable flaw. Understanding that pattern explains why some options still work today and others quietly stopped years ago.
What Is Active Probing?
Active probing is the firewall's confirmation step. Passive detection (reading packet patterns) can flag a connection as suspicious, but it doesn't prove what's on the other end. So separate infrastructure actually connects to the suspected server itself, sending test messages designed to see how a VPN server specifically would respond, before committing to a block. One detailed study of this against Shadowsocks servers logged over 51,000 active probes from more than 12,000 distinct Chinese IP addresses, arriving anywhere from seconds to weeks after the original flagged connection, with timing patterns suggesting a small number of centrally coordinated systems doing the probing rather than random scanning. A separate, award-winning 2022 study found OpenVPN's handshake fingerprintable with over 85% accuracy in real internet service provider testing, because OpenVPN's connection setup follows a fixed, recognizable sequence with an unencrypted header. WireGuard has a comparable weakness: its handshake messages have fixed types and lengths, documented directly on WireGuard's own developer mailing list, which is compact and clean enough to also risk tripping the entropy rule described above.
Shadowsocks, an early and once-popular circumvention tool, relied purely on high-entropy encryption to avoid signature matching, and was vulnerable both to active probing against weak ciphers and to the 2021 entropy rule once that arrived. The current strongest approach, a combination called VLESS with XTLS and REALITY, takes a fundamentally different strategy: rather than disguising a VPN handshake, it makes the server impersonate the real TLS handshake of an actual, unrelated website, such as a major CDN-hosted domain, without owning that site's certificate. Any connection that doesn't authenticate correctly, including a firewall probe, gets silently forwarded to the real site instead, so the prober sees exactly what a genuine visit to that website looks like. There's nothing to catch, because nothing about the disguise is fake at the point where a probe would look. Trojan works on a similar principle, running genuine TLS with a real certificate and showing an ordinary decoy website to anyone who connects without the right credentials.
The pattern across all of this is a shift in strategy. Early tools tried to disguise a VPN's traffic. Current tools try to eliminate anything distinguishing at all by borrowing a real website's identity outright. That's a meaningfully harder thing to detect, but it's not unbeatable. Documented reports from Russia in late 2025 describe REALITY connections being blocked through active tampering with the handshake itself, and testing from Iran found that REALITY's survival depends heavily on how much traffic a server carries and how fresh its IP address is: a high-traffic server was blocked within two hours in one test, while a near-idle one on the same protocol survived over a week. Protocol choice matters. It is not the only thing that matters.
Do VPNs Actually Work in China?
Sometimes, for a while, depending on the protocol, the provider, the current political calendar, and how much traffic the server is carrying. That's an honest answer, not a hedge. Reliability has been documented to drop noticeably around politically sensitive periods, including Communist Party congresses and the anniversary of the Tiananmen Square crackdown, when filtering intensifies for a window of days to weeks. No commercial VPN has a clean, permanent track record. Even long-standing, well-regarded providers have had extended total outages: one popular option marketed on its reliability was reported down for close to two months in early 2026, according to travel guides tracking the outage.
What to Actually Look For
Rather than chasing a specific brand name that may already be unreliable by the time you read this, these are the criteria that hold up regardless of which provider is having a good month:
Built-in obfuscation, not a bare default protocol. A provider using REALITY, Trojan, or a similarly disguised connection method is working with the current strongest approach; a provider defaulting to plain OpenVPN or IKEv2 is using something with a well-documented fingerprint.
A real multi-year track record among long-term residents and expats, not just marketing copy claiming the service "works in China."
Transparent ownership. Know who actually runs the service and where the company is based. This matters more than almost anything else, for reasons covered in the next section.
Honesty about impermanence. A provider willing to say plainly that nothing works forever there is more trustworthy than one promising guaranteed, permanent access.
A self-hosted option, for technically inclined users. Running your own server on a personal VPS outside China, rather than a shared commercial VPN IP, tends to last longer simply because it isn't already flagged as VPN infrastructure at scale, though it requires ongoing setup and maintenance rather than a one-time purchase.
Is Using a VPN in China Illegal?
Technically, this sits in a genuine gray area, and the honest answer requires holding two facts at once. A January 2017 regulation from China's Ministry of Industry and Information Technology requires telecom-regulator approval for any cross-border dedicated line, including VPN services, and its text is broad enough to cover individual use if read literally. But MIIT itself publicly clarified the same year that "there is no ban on VPN services to individuals," stating the rule targets unlicensed commercial providers and resellers, not personal or foreign-company use. That's a real, quotable gap between what the regulation technically says and what the regulator has said it's actually for.
Documented enforcement against individuals is real, but small and rare. A man in Guangdong was fined 1,000 yuan (about $150) in 2018 for using a VPN. A man in Shaanxi was fined 500 yuan in 2020, a case that drew enough public pushback that a state media editor publicly argued the fine was excessive. Zhejiang province recorded roughly 62 individual cases over a 15-month window, a very small number for one of China's most internet-connected provinces. These are administrative penalties, not criminal charges, and no documented case of a foreign tourist or business traveler being fined or detained for personal VPN use turned up in researching this article. Selling VPN access is a different matter entirely: individuals convicted of commercially reselling VPN service have received actual prison sentences, in one case 5.5 years plus a fine of 500,000 yuan, showing where the real enforcement weight falls.
Is China Cracking Down Harder on VPNs in 2026?
The trend line points that way, though the newest reporting should be read with some caution. Leaked internal documents reportedly circulating since April 2026, describing coordinated instructions to telecom carriers to terminate circumvention services, have been covered by outlets including China Digital Times, which itself frames the documents as unverified rather than confirmed policy. What is independently documented is the technical trend behind it: expanded QUIC-based filtering since 2024, and a discovery that at least one province, Henan, has been running its own separate, larger censorship layer since 2023. The practical risk of a VPN simply failing to work appears to be rising faster than the legal risk to any individual using one.
The Real Risk Isn't the Firewall, It's the Fake VPN App
This is the part worth taking seriously before installing anything. Independent research found that roughly one in five of the top 100 free VPN apps in the US App Store, representing more than 70 million combined downloads, are secretly owned by Chinese companies, several tied to Qihoo 360, including apps marketed as Turbo VPN, VPN Proxy Master, Thunder VPN, Snap VPN, and Signal Secure VPN, according to Cybernews and corroborated by Malwarebytes. Under China's National Intelligence Law, Chinese-domiciled companies are legally obligated to cooperate with state security services on data access. A "free" VPN routed through a company with that obligation doesn't just carry the usual free-app risks of ad injection and tracking, it can defeat the entire point of using a VPN in a country whose internet is already closely monitored, since privacy from the state is specifically what you're trying to get.
If You're Traveling to or Working in China
Install and configure your VPN before you arrive. VPN provider websites and app stores are generally unreachable from inside China, so getting there without one already set up is a real problem, not a minor inconvenience. This isn't new: Apple removed major VPN apps, including ExpressVPN, from China's App Store back in July 2017 in response to the regulatory push, and the pattern of apps simply not being available locally has continued since.
Carry a backup provider, since protocol-specific and provider-specific blocks aren't evenly distributed and don't always announce themselves. Expect connections to slow down or drop around major political anniversaries. If you're traveling for sensitive work, consider a secondary device with minimal personal or proprietary data loaded onto it, consistent with general guidance for travel to countries where telecom providers can be compelled to grant government access to networks and data.
FAQ
Is it illegal to use a VPN in China?
Technically it falls under a regulation requiring approval for cross-border connections, but China's own telecom regulator has publicly stated the rule targets unlicensed commercial providers, not individuals. Documented penalties against individuals exist but are small, rare, and administrative rather than criminal, and no confirmed case involves a foreign tourist or business traveler being penalized for personal use.
Is a free VPN safe to use in China?
Not reliably. A meaningful share of popular free VPN apps have undisclosed ownership tied to companies obligated to cooperate with government data requests in their home jurisdiction, on top of the usual free-app risks like ad injection and tracking. Checking a VPN publisher before installing is a reasonable precaution, not paranoia.
Will my VPN just get slower before it stops working entirely?
Sometimes, though this isn't as formally documented as the firewall's blocking mechanisms. Reported slowdowns tend to coincide with broader network congestion during politically sensitive periods rather than being a distinct, deliberate throttling feature separate from outright blocking.
Does using an eSIM instead of a local SIM change any of this?
No. An eSIM only affects how your device gets cellular data access, not how that traffic is inspected once it reaches the network. VPN detection happens at the carrier and gateway level regardless of whether your data connection comes from a physical SIM or an eSIM.
If you're about to install a VPN for a trip to China, or you're already using one and something about it feels off, run the app's publisher or the provider's domain through ScamInfo's validator before you trust it with your traffic. If you think you've already installed something that's quietly harvesting your data instead of protecting it, report it so the pattern can be flagged for other users.