Start your 3-day free trial
Sign up to experience all premium features at no cost.
*Available only to new users. Each user is limited to one trial.


Encrypted Client Hello (ECH) can prevent a network observer from reading the real server name in a successfully negotiated TLS ClientHelloInner. It does not guarantee access: deployment requires a compatible client, a usable ECH configuration, and server infrastructure. That configuration is commonly discovered through DNS but can also be preconfigured, while destination IP addresses, connection timing, volume, the outer ClientHello, and recognizable non-TLS protocols can remain visible.
The complete VPN guide explains where a VPN tunnel sits. ECH is narrower: it protects sensitive TLS handshake metadata between an ECH-capable client and server. It is not a VPN and does not encrypt the entire network path by itself.
Key Takeaways
- ECH encrypts a ClientHelloInner that can contain the real SNI and other sensitive extensions.
- The visible ClientHelloOuter is designed to be safe for public exposure and carries the ECH extension.
- The client needs a usable, supported ECH configuration, commonly discovered through HTTPS/SVCB DNS records or preconfigured.
- Rejection and retry behavior are part of the protocol, so “ECH enabled” is not the same as “ECH accepted.”
- IP, traffic-analysis, endpoint, and protocol rules remain possible even when SNI is protected.
Traditional TLS Server Name Indication places the requested hostname in the server_name extension of the ClientHello. RFC 6066 defined SNI so a server hosting several names on one address could select appropriate credentials before the encrypted application request existed.[3] The extension improved virtual hosting but exposed the name to passive observers.
RFC 9849 standardizes TLS Encrypted Client Hello. The client constructs an inner ClientHello with sensitive values and encrypts it to a public key from an ECH configuration. It also sends an outer ClientHello that can be processed by client-facing infrastructure.[1] When the server accepts ECH, the inner handshake becomes the basis of the TLS connection.
ECH protects more than the literal SNI field because multiple sensitive extensions can live in the inner message. Its core privacy result is still bounded: it conceals those inner values from an observer that cannot decrypt them. It does not conceal the fact that a connection exists.
An access network still needs routable information. The destination and source IP addresses remain in packet headers. Transport details, packet sizes, timing, connection duration, direction, and total volume are observable. The network may also see DNS traffic unless it is protected through a separate mechanism.
The outer ClientHello remains visible. RFC 9849 defines consistency rules between inner and outer messages and mechanisms for public-name deployment.[1] ECH aims to avoid exposing the sensitive origin name, but it does not make every outer byte indistinguishable across all implementations.
After a successful TLS handshake, application data is encrypted by TLS. Traffic analysis can still correlate endpoints or patterns without reading the HTTP path or inner hostname. What an ISP can see when you use a VPN applies the same evidence discipline to a different encrypted outer flow.
ECH also needs support on both sides: in your browser and at the destination. When a site publishes no ECH configuration and your concern is the local network reading its name, a VPN covers that at a different layer: for traffic routed into the tunnel, the local network sees the VPN connection while the site's handshake and destination address travel inside it. Bypassed traffic and DNS sent outside the tunnel retain their own visibility.
With AethoVPN, connect to a location in the app before opening the site, then compare what the local network can observe with your ECH-only attempt. For that tunneled traffic, the name and address may still be observable beyond the VPN exit according to the destination connection's own protections, and the tunnel is still a visible connection that a network can choose to block. After checking local law and the network's rules, start the 3-day free trial and open the same site with and without the tunnel.
The client needs an ECHConfigList containing parameters and a public key. RFC 9849 describes the TLS processing, while RFC 9848 defines how the ech service parameter in SVCB and HTTPS resource records distributes ECH configuration.[1][2]
Initial discovery does not always authenticate an ECH configuration. RFC 9849 permits HTTPS/SVCB delivery without verifiable authenticity or provenance, while preconfiguration and authenticated DNS can provide different assurances. The client still follows its DNS and TLS security model; protected DNS can reduce exposure and tampering on that path, but encrypted DNS is a separate protocol with separate deployment.[1][2]
Freshness matters. Servers can rotate ECH keys and publish updated configurations. Cached or stale data may cause rejection and a retry with authenticated replacement configuration. That recovery path must not train the client to silently discard security whenever an on-path observer interferes.
On acceptance, client and server use the ClientHelloInner for the handshake, subject to RFC 9849’s confirmation and transcript rules.[1] The sensitive name is not sent in clear as traditional SNI. The outer message remains the public-facing envelope.
If the server cannot decrypt or accept the offer, it can reject ECH and provide retry configurations through the standardized path. The client must authenticate public_name and complete the outer TLS handshake before trusting the retry configurations. The rejected connection must then be closed with ech_required before any application data is sent; a supported retry configuration can be used for a retry on a new transport connection. If authentication or the handshake fails, the client must fail rather than trust the retry configurations or treat the failure as a secure signal to disable ECH.[1]
Failure does not have one appearance. A stale configuration can cause a recoverable retry. A misconfigured server can fail TLS. A network can drop the first ClientHello, interfere with DNS discovery, block the destination address, or allow the exchange. Client telemetry should distinguish offered, accepted, rejected, retried, and ordinary non-ECH states.
| Branch | Inner name protected from passive path observer? | What the observer can still see | Result boundary |
|---|---|---|---|
| ECH offered and accepted | Yes, assuming ECH cryptography and endpoints are not compromised | IPs, transport, timing, sizes, outer ClientHello | TLS can continue using the inner message |
| ECH rejected with authenticated retry data | The original inner remains encrypted | Rejection-related traffic and visible metadata | Client may retry with a valid new configuration |
| ECH unavailable | No ECH protection applies | Traditional ClientHello fields may be visible | Client policy decides whether ordinary TLS is allowed |
| Connection dropped before completion | The encrypted inner may remain unreadable | Destination and traffic pattern | No proof of why or where the drop occurred |
ECH can defeat a rule whose only usable selector is the cleartext inner server name, provided the client obtains valid configuration and the server accepts the encrypted ClientHello. In that narrow case, the observer no longer receives the traditional SNI value it expected to match.
That does not mean the service is unblocking-proof. A policy can select a destination IP or prefix, block all traffic to a public-facing endpoint, restrict ports, detect a distinct protocol, or deny connections it cannot classify. Shared infrastructure may make address blocking costly, but collateral cost is a policy consideration rather than a technical guarantee.
The website-blocking overview compares other selectors. ECH changes the visibility of TLS metadata; it does not remove the name from the server’s own application context, change authorization, or force an access network to carry the connection.
ECH moves sensitive extensions into the inner message, so a passive observer cannot build a fingerprint from values it can no longer read. The outer ClientHello still has structure, supported parameters, lengths, and implementation choices. Packet-level timing and size features also remain.
TLS fingerprinting is classification based on visible characteristics. ECH can change or reduce the available feature set, but it does not make all clients emit identical traffic. Padding and careful outer construction help privacy but are not a promise of universal indistinguishability.
Classification is not identity. A fingerprint match can produce false positives and false negatives, while an encrypted inner handshake can still be associated with a known destination address. Treat “ECH hides SNI” and “ECH hides every TLS or network feature” as separate claims.
No. ECH is a TLS extension used between a TLS client and ECH-capable server infrastructure. A VPN usually creates a protected path between a device and VPN endpoint and may carry many applications inside that tunnel. VPN versus HTTPS explains the layer difference.
Some VPN products may use TLS in a control plane or transport, but that does not imply they implement ECH for every VPN connection. Conversely, a browser can use ECH for HTTPS without a VPN. Product labels should not be used to infer protocol negotiation.
Confirm ECH at the actual client-server handshake, and treat any VPN protection as a separate layer that neither enables ECH nor guarantees access through a blocking policy.
Begin with compatibility: identify the client version, server deployment, and how the client obtains its ECH configuration. For DNS discovery, check the resolver path, target HTTPS record, and advertised ECH configuration. For preconfigured clients, check the ECHConfigList actually loaded and whether it contains a configuration supported by the client and compatible with the server deployment; an absent HTTPS record alone does not rule out ECH.[1] In either case, use browser or application diagnostics designed to report ECH status. A generic padlock proves TLS, not ECH acceptance.
If you administer both ends, correlate the client’s offered/accepted state with server telemetry for the same timestamp. Confirm that the expected inner name was used and that the certificate and application origin were correct. Do not weaken certificate validation to make an experiment pass.
Separate privacy from reachability. An accepted ECH handshake supports the claim that the inner ClientHello was protected for that connection. It does not prove that the network learned nothing, that future connections use ECH, or that another destination cannot be blocked by IP or behavior.
No. Routers need the destination address to deliver packets. ECH protects content inside the TLS ClientHello, not the IP header.
Only when valid ECH configuration is used and the server accepts the encrypted inner handshake. DNS, application behavior, destination infrastructure, or fallback can expose other clues.
ECH protects the sensitive server name by placing it in ClientHelloInner while retaining an outer handshake for deployment. Servers still need a name to select the intended service.
No. Encrypted DNS can protect the DNS exchange, while ECH requires a compatible client, a usable ECH configuration obtained through DNS or preconfiguration, and accepting server infrastructure. They solve related but separate problems.
A network can drop connections, addresses, ports, or traffic it classifies as ECH-related. Whether that is practical or creates collateral effects depends on the deployment and policy.
No. It removes sensitive inner fields from passive view, but the outer handshake and traffic metadata can still vary by implementation, destination, and session.
Use client and server diagnostics that explicitly report the ECH negotiation state. A successful HTTPS page or valid certificate alone does not prove ECH acceptance.
Disclaimer: This article provides general technical information. Network behavior and client support change over time; follow applicable policy and do not disable certificate or DNS security checks during testing.
Sources:
Sources checked 9 October 2026.
Related Articles:
Sign up to experience all premium features at no cost.
*Available only to new users. Each user is limited to one trial.