Encrypted DNS: DoH, DoT, and DoQ
1. Overview
Encrypted DNS refers to a family of standards (DoT, DoH, DoQ, ODoH) that carry DNS queries and responses—traditionally exchanged in plaintext over UDP/TCP port 53—inside a transport channel encrypted with TLS or QUIC, thereby blocking eavesdropping and tampering along the path. Whereas DNSSEC targets the origin authentication and integrity of response data, encrypted DNS targets the confidentiality (privacy) and transport-segment protection of the query itself, so the two differ in their axis of purpose.
The emergence of encrypted DNS is rooted in the fact that DNS was the last core protocol on the internet to remain in plaintext. Web traffic moved rapidly to HTTPS, mail to STARTTLS, and the transport layer to TLS 1.3, yet the DNS query—which plainly reveals "which site am I trying to reach"—long retained the plaintext short-message structure of its 1983 design. As a result, the moment a user looked up www.hospital.example, that name became readable metadata for the subscriber-network ISP, the administrator of a public Wi-Fi, or any on-path intermediary. Even with HTTPS hiding the payload, "where you connected" leaked out through DNS and the SNI in the TLS ClientHello, and this asymmetry became the direct motivation for encrypted DNS.
The need becomes clear in three directions. First is privacy protection, structurally blocking a user's browsing history from becoming fodder for ISP logs, ad profiling, or state-level surveillance. Second is integrity and security, using transport encryption and authentication to stop cache poisoning aimed at plaintext DNS, on-path response forgery, and DNS-based censorship and hijacking that selectively redirects specific domains to fake IPs. Third is the shift of the trust model, turning "a query visible to everyone on the path" into "a query shared only with the chosen resolver." That third point, however, gives rise to the new issue of "trust concentration on the resolver" discussed later.
In practice, the risk on untrusted networks such as public Wi-Fi, hotels, and airports makes this need tangible. An attacker on the same network can intercept plaintext DNS responses to lure a user to a fake login page (pharming) or redirect a financial domain to an attacker's server. Encrypted DNS blocks such local-network-level man-in-the-middle attacks with transport encryption and resolver authentication, so its benefit grows the more mobile the modern usage environment becomes. In this sense, encrypted DNS has a dual character as both a "privacy technology" and an "anti-phishing/anti-pharming technology."
The essential characteristics that define encrypted DNS can be summarized as transport-layer protection, complementarity with DNSSEC, and a shift in the trusting party. Because it hides the transport segment with TLS/QUIC, it does not verify the cached data itself (DNSSEC handles that role), so the two are complementary rather than competing. It also moves "who sees the query" from the ISP to a specific resolver operator, which is the key to this technology's double-edged nature—embracing both technical gains and governance issues.
The standardization trajectory also reflects and evolves with this duality. In 2016, the dedicated-port DoT (RFC 7858) came first and offered "manageable encryption"; in 2018, DoH (RFC 8484), which blends into web traffic, sparked debate as "censorship-resistant encryption"; and in 2022, the low-latency DoQ (RFC 9250) and the sender-query-separating ODoH (RFC 9230) were standardized side by side in the same year. Rather than one technology replacing the next, the hallmark of encrypted DNS's standards history is that methods with different trust models and performance goals coexist and widen the range of choices.
2. Overall Architecture and Threat Model of Encrypted DNS
The key to understanding encrypted DNS is its segment-limited nature: it wraps the last mile between the stub resolver (endpoint) and the recursive resolver (public DNS) in an encrypted tunnel. The upstream segment toward the authoritative servers is often still plaintext, so what encrypted DNS hides is the endpoint-to-resolver segment, where "what the user asked" is exposed most sensitively. The overall diagram below contrasts the exposure of plaintext DNS with the segment that encrypted DNS protects.
flowchart TB
subgraph PLAIN["Exposure of plaintext DNS (Do53)"]
U1["Endpoint stub resolver"] -- "Plaintext UDP/TCP 53" --> R1["Recursive resolver"]
MITM["On-path ISP/intermediary: can read/alter queries"]
U1 -.exposed.-> MITM
end
subgraph ENC["Segment protected by encrypted DNS"]
U2["Endpoint stub resolver"] -- "TLS/QUIC encrypted tunnel" --> R2["Recursive resolver (DoT/DoH/DoQ)"]
R2 -- "Plaintext or encrypted (upstream)" --> AUTH["Authoritative name server"]
end
From a threat-model standpoint, it is important to distinguish what encrypted DNS blocks from what it does not. It blocks passive eavesdropping (collecting who looks up what) and active tampering (response forgery/injection) on the endpoint-to-resolver segment. What it does not block is the query being read by the chosen resolver operator itself (the resolver sees the query in plaintext), and after-the-fact inference via the destination IP and TLS SNI. For example, even if DoH hides DNS, if the SNI in the subsequent TLS handshake's ClientHello is plaintext, "which site" is still revealed. For this reason, connection privacy is completed only when encrypted DNS is paired with ECH (Encrypted Client Hello); an answer must note that either one alone is only half the solution.
One more point to note is that what encrypted DNS protects is limited to the "endpoint-to-resolver" segment. The upstream queries the recursive resolver sends while climbing the authoritative name server chain may still be plaintext, and the authoritative servers in this segment see part of the query. This is complemented by QNAME Minimization (RFC 9156), a technique whereby the resolver exposes to each authoritative server only the labels strictly needed at that step (e.g., only com to the root, only example.com to the TLD) instead of the full domain (a.b.example.com). If encrypted DNS hides the query from "third parties on the path," QNAME minimization minimizes exposure to "upstream authoritative servers," so both must be applied together for DNS privacy to be completed end to end. In an exam answer, the discriminating insight is to view encrypted DNS not as a single technology but as a privacy stack interwoven with DNSSEC, ECH, and QNAME minimization.
sequenceDiagram
participant S as Endpoint stub
participant R as Recursive resolver
participant A as Authoritative name server
S->>R: Establish TLS/QUIC session (certificate validation)
S->>R: Encrypted query (QNAME)
R->>A: Iterative queries
A-->>R: Response (RRset)
R-->>S: Encrypted response
Note over S,R: On-path intermediaries cannot see session contents
As the flow above shows, the security effect of encrypted DNS hinges entirely on the establishment and authentication of the stub-to-resolver session. Once the session is established, subsequent queries reuse the same tunnel (keep-alive) to amortize handshake cost, and at that point the resolver processes the decrypted plaintext query. This is precisely the moment where the trust boundary of encrypted DNS narrows from "third parties on the path" to "the resolver operator." Therefore, which resolver to designate as the trust endpoint (a public global operator, a trusted domestic operator, or self-operated) becomes a governance decision as important as the technology choice itself.
3. Main Methods: DoT, DoH, DoQ, and ODoH
Encrypted DNS methods diverge by "what wraps it, which port it uses, and how it blends in." Each method chose a different point on the tension of management visibility vs. user privacy. Below, each of the four methods is examined with its differences in standard, port, transport, and trust model, after which selection criteria are synthesized.
The common security premise for all four methods is resolver authentication. If one merely builds an encrypted tunnel without verifying who the counterpart is, an intermediary can impersonate the resolver, so each method validates the resolver's TLS certificate. Here the trust level splits into two profiles. Strict mode mandatorily validates the resolver's name and certificate and rejects the query on failure, fundamentally blocking downgrade; Opportunistic mode uses encryption when possible but falls back to plaintext on failure, prioritizing availability. The higher the security grade of an environment, the more it favors strict mode, while the more compatibility-sensitive an environment is, the more it favors opportunistic mode—a clear trade-off.
A. DoT (DNS over TLS, RFC 7858)
DoT carries DNS messages directly over a TLS session on dedicated port 853 (standardized in 2016). Because the structure is simple and the port is separated, enterprises and carriers gain the advantage that DNS traffic is easy to identify and manage. Policy can be applied simply by opening or closing port 853 at the firewall, and security monitoring tools can distinguish and trace the DNS flow. As Android supported DoT at the system level via its "Private DNS" feature, it spread relatively widely in mobile environments. The dedicated port also lets resolver operators run a DNS-only TLS server, giving a benefit of clear operations and capacity planning. Conversely, a drawback is that, being clearly distinct from plaintext Do53, the very fact of communication is not hidden, so the "unobservability" of privacy is lower than with DoH.
Yet that very "ease of identification" is a weakness from a privacy standpoint. The existence of traffic on port 853 itself reveals that "this user is using encrypted DNS," and in censorship environments it becomes an easy target to block port 853 wholesale. In fact, some networks block 853 to disable DoT and fall back to plaintext DNS. In short, DoT sits at a "management-friendly but easily blocked/identified" compromise point.
B. DoH (DNS over HTTPS, RFC 8484)
DoH carries DNS queries inside an HTTPS request (the /dns-query path) over the same port 443 as ordinary web traffic (standardized in 2018). The key difference is that DoH traffic is nearly impossible to distinguish, blended in with all other HTTPS traffic. It is the strongest in terms of privacy and censorship resistance, since it works as long as 443 is open and selectively blocking only DNS is practically impossible. It became mainstream through browsers as Firefox shipped "DoH by default" and Chrome shipped "Secure DNS."
This very strength leads to a severe loss of visibility from an enterprise security standpoint. Traditionally, organizations have detected malicious-domain access, data exfiltration (DNS tunneling), and malware C2 communication from DNS logs, but if an application bypasses the operating system's DNS settings to query an external DoH resolver directly, all of these controls are nullified. Hence many enterprises respond by blocking known public DoH endpoints, or by using browser enterprise policy to disable DoH and pin it to an internal resolver. DoH is a stark example of the two extremes of "the best privacy for the user, the greatest loss of control for the administrator."
Meanwhile, DoH has a subtle weakness arising from its transport method. Because it places the DNS query in an HTTP request, the per-message overhead is larger than Do53, with the cost of HTTP header/method (GET/POST) processing added on. Also, the characteristic that DoH traffic blends into ordinary HTTPS can paradoxically become the target of traffic-analysis (packet-size/timing-pattern) identification attempts. That is, "even if it cannot be blocked by port, it may be inferred by statistical fingerprint," so it does not guarantee complete anonymity, and padding to flatten the message-length distribution is discussed as a complementary measure.
C. DoQ (DNS over QUIC, RFC 9250)
DoQ maps DNS directly onto QUIC (UDP-based) transport; it was standardized in 2022 and uses UDP port 853 (port 53 is prohibited). QUIC embeds TLS 1.3 and has the characteristics of greatly reducing connection-setup latency (0-RTT/1-RTT) and keeping other streams unaffected by packet loss (avoiding HoL blocking). Whereas DoT is bound by TCP's head-of-line delay and TLS handshake round trips, DoQ's strength, as touted by the standard, is that it provides privacy on par with DoT while achieving low latency close to plaintext UDP DNS. Since each query is handled as a separate stream, parallelism across many queries is also good. Unlike DoH over HTTP/3, which gains only "part" of QUIC's benefits, DoQ's design intent is to be lighter by mapping directly onto QUIC. However, because it uses the dedicated UDP port 853, it sits in a position similar to DoT in terms of ease of identification/blocking.
The areas where DoQ draws particular attention are the recursive-to-authoritative segment and mobile. QUIC's connection migration maintains the session even when the endpoint's IP changes (Wi-Fi↔cellular handoff), reducing re-handshake cost in environments with frequent movement. Being UDP-based, its transport characteristics are also similar to plaintext DNS, so it is considered a high-performance encrypted-transport alternative for large-scale resolver-to-authoritative communication. On the other hand, it cannot operate on conservative networks that block UDP, so a fallback design to TCP-based DoT/DoH is an operational consideration.
D. ODoH (Oblivious DoH, RFC 9230)
The three preceding methods all share the limitation that "the chosen resolver sees both 'my IP' and 'my query' at once." ODoH resolves this by separating the proxy and the target. The user encrypts the query content with the target resolver's public key and sends it to a relay proxy, which forwards only the ciphertext to the target while hiding the user's IP. As a result, the proxy knows only "who sent it" but not the query content, while the target knows only "what is being asked" but not the sender. Unless the two collude, neither can combine "IP + query," which structurally mitigates the trust-concentration problem on the resolver.
This separation principle is simple, but the trust assumption is key. ODoH's privacy stands on the non-collusion assumption that "the proxy and the target do not collude," so it is meaningful only when the two are placed under different independent operators/jurisdictions. If the same operator runs both, the benefit of separation disappears. Apple's iCloud Private Relay, which entrusts the "ingress proxy (user identification)" and the "egress proxy (destination identification)" to different parties to separate the two, follows the same design philosophy. However, with the added relay hop increasing latency and operational complexity, and the practical constraint of securing both proxy and target as trustworthy independent operators, it has not yet reached a popular default. The data flow of ODoH can be diagrammed as follows.
flowchart LR
C["User: encrypts query with target public key"] -- "Ciphertext + user IP" --> P["Relay proxy: sees IP but not content"]
P -- "Ciphertext only (IP hidden)" --> T["Target resolver: sees content but not sender"]
T -- "Encrypted response" --> P
P -- "Encrypted response" --> C
| Category | DoT | DoH | DoQ | ODoH |
|---|---|---|---|---|
| Standard | RFC 7858 | RFC 8484 | RFC 9250 | RFC 9230 |
| Transport | TLS/TCP | HTTPS/TCP·H3 | QUIC/UDP | HTTPS (dual hop) |
| Port | 853 | 443 | 853 (UDP) | 443 |
| Identify/block | Easy | Very hard | Easy | Hard |
| Management visibility | High | Low | High | Low |
| Key strength | Easy to manage | Censorship-resistant | Low-latency·parallel | IP/query separation |
Note that the "identify/block" and "management visibility" rows in the table above essentially move in opposite directions along a single spectrum. High management visibility means traffic is easy to identify and block, and strong censorship resistance means the organization's control is also hard. On the premise of accepting this trade-off, the practical design principle is to combine and pin methods according to the nature of the network (public vs. enterprise) and what is being protected.
4. Adoption/Discovery Mechanisms and Comparison
The actual spread of encrypted DNS entails a discovery problem: "how does the endpoint find out which encrypted resolver to use?" Early on, browsers/OSes often hardcoded a public resolver (e.g., a specific public DNS), which drew backlash for ignoring the network operator's intent (split-horizon DNS, internal-domain resolution) and diverting queries outside. To mitigate this, standards such as DDR (Discovery of Designated Resolvers, RFC 9462) and DNR (Discovery of Network-designated Resolvers, RFC 9463) were established so that the network informs the endpoint of the existence of its designated encrypted resolver, helping a natural upgrade from plaintext to an encrypted tunnel. This created a trend toward reconciling "user privacy" with "the network operator's legitimate DNS control."
The fundamental reason the three methods differ lies in the design choice of "how much the encrypted DNS traffic is blended with other traffic." DoT/DoQ, which use a dedicated port (853), have conspicuous traffic, so they favor enterprise management but are vulnerable to censorship/blocking; DoH, which melts into web traffic, is the opposite. The practical implication is clear. For individual users where autonomy and censorship resistance matter, DoH (+ECH) is suitable; for enterprise internal networks where log-based security detection is essential, "DoT/DoH pinned to an internal resolver" is suitable. That is, no method is "superior"; rather, the optimal point shifts depending on what asset and trust boundary you are protecting, which is the core of the comparison.
Another important comparison axis is the degree of trust concentration. DoT, DoH, and DoQ all concentrate trust in one party in that "a single chosen resolver sees the query + IP," whereas ODoH distributes it to a proxy and a target. Therefore, in sensitive fields (medical, financial, public), to structurally lower "the risk of the resolver operator accumulating a user's browsing history," one must weigh a separated (oblivious) architecture beyond simple transport encryption. This is a threat-model-level decision of "is hiding only the transport segment enough, or should the resolver itself be regarded as untrusted?"—showing that choosing an encrypted DNS method is not a mere performance comparison but a choice of trust assumptions.
As a concrete case, one global browser, when enabling DoH by default in the US, first detected whether the network performs content filtering by looking up a "canary domain," designing it so that DoH is not turned on if home filtering or a managed enterprise environment is detected. This is a representative example showing that technology adoption is not a pure technical problem but a governance compromise among home protection, enterprise control, and user privacy. Domestically as well, there was debate over how the method of blocking specific sites by intercepting the SNI field was nullified by the spread of ESNI/ECH, which reveals that encrypted DNS/ECH is a technology that collides head-on with censorship infrastructure.
A performance comparison also influences practical choices. Plaintext Do53 usually gets a response within 1 RTT but has no protection, while DoT adds several round trips of delay on the first connection from the TCP 3-way handshake + TLS handshake. DoH has HTTP-layer overhead but absorbs much of it through connection reuse (keep-alive) and HTTP/2 multiplexing. DoQ, thanks to QUIC's 1-RTT (0-RTT on reconnection) setup and stream independence, is particularly advantageous in mobile/wireless environments with high packet-loss rates. For example, on a lossy cellular network, DoT has a single packet loss delay the entire TCP stream, whereas DoQ retransmits only the affected stream, reducing perceived latency. However, since the initial handshake cost is distributed through caching and connection keep-alive, the actual perceived difference for users hinges more on the "connection-reuse strategy" than on the "first time," which must be considered together.
5. Deep Dive: Enterprise Security Visibility, Standards, and Trends
This section addresses the deep issues encrypted DNS poses for security operations and the standards ecosystem. The biggest challenge encrypted DNS poses to security practice is the "loss of DNS visibility." Traditional security operations centers (SOCs) have used DNS query logs as a core signal for malware C2 detection, data exfiltration via DNS tunneling, and blocking newly registered malicious domains (NRDs). Yet if an endpoint application bypasses OS settings to query an external DoH directly, this signal vanishes entirely. Practical responses converge on three branches. First, endpoint policy disables DoH in the browser/OS or forcibly pins it to an internal encrypted resolver operated by the organization. Second, network policy blocks known public DoH/DoT endpoints and induces upgrade by "designating" the organizational resolver via DDR/DNR. Third, a Protective DNS service is placed internally so the organization holds the termination of the encrypted tunnel and performs threat-intelligence-based filtering and logging at that point. National cybersecurity agencies are also increasingly recommending the operation of a "Protective DNS" for organizations.
From a DNS-tunneling-detection standpoint, encrypted DNS is also a double-edged sword. Attackers encode data into the subdomain labels of DNS queries to bypass firewalls via DNS tunneling, exfiltrating data or hiding C2 communication; in the plaintext DNS era, this was detected by analyzing query length, entropy, frequency, and the like from logs. Encrypted DNS hides this very analysis target, so an organization must move the detection point from network logs to the termination resolver/endpoint agent. That is, the detection architecture must be redesigned so that the organization owns the "termination" of the encrypted tunnel and analyzes the decrypted query there, or an EDR captures abnormal DNS calls at the process level. This is a representative case showing the structural tension that "the more encryption raises privacy, the more it lowers the defender's visibility."
In terms of standards and technology trends, three streams stand out. First is the maturation of ECH (Encrypted Client Hello). Because the destination is revealed if the subsequent TLS SNI is plaintext even after encrypted DNS hides the query, ECH—which encrypts even the SNI—must spread for connection privacy to be completed. Second is the rise of DoQ and HTTP/3-based DoH, where QUIC's advantages of low latency and connection migration boost DNS performance in mobile environments. Third is the institutionalization of sender-query separation (the ODoH family) and relay-based privacy, continuing efforts to cryptographically minimize trust in the resolver operator. In the long run, as QUIC/TLS 1.3 transition to PQC (post-quantum cryptography) key exchange, the session establishment of encrypted DNS is also expected to migrate to PQC hybrids.
There are also significant implications from a domestic (Korean) perspective. The harmful-information blocking regime that had operated on SNI-based blocking is gradually losing technical effectiveness with the spread of ECH and encrypted DNS, so social consensus between blocking policy and user privacy/net neutrality emerges as a task. At the same time, since reliance on external global resolvers may be undesirable for public and financial institutions in terms of data sovereignty and auditability, a strategy of securing a trustworthy encrypted resolver on a self-operated or domestic-operator basis is required. In other words, encrypted DNS is both a matter of technology adoption and a matter of digital sovereignty and regulatory design.
6. Considerations and Implications
Complementary design (DNSSEC + Encrypted DNS + ECH + QNAME minimization): Encrypted DNS handles transport confidentiality, DNSSEC handles data integrity/origin authentication, ECH handles SNI concealment, and QNAME minimization handles minimizing upstream authoritative-server exposure. Since no single one is complete in either privacy or integrity, a professional engineer must design and recommend these as a layered combination rather than a single control. Trusting only encryption while neglecting integrity leaves one directly exposed to response forgery by a malicious resolver, and DNSSEC alone lets the query content leak in plaintext. Clarifying the "division of roles" whereby each technology handles a different threat is the starting point of the design.
Redefining the trust boundary and governance trade-offs: Encrypted DNS merely shifts the party reading the query from the ISP to a specific resolver operator—it does not eliminate it. Concentration in a handful of global resolvers raises concerns of a single point of failure/censorship and data monopoly, so sensitive fields such as public and financial sectors should examine self-operation or a trusted domestic resolver and an ODoH-style separated architecture. A perspective of guarding against privacy enhancement ending up as deepened centralization is needed.
Balancing security visibility and privacy: In enterprise environments, "user protection" and "visibility for threat detection" collide. Rather than wholesale blocking DoH, a design of "keep encryption but let the organization hold the control point" via upgrade to the organizational resolver (DDR/DNR) + Protective DNS + termination logging is the realistic solution. However, since this presupposes "trust in the internal resolver," strengthening that resolver's security/audit regime must come first.
Performance/availability/fallback risk: One must jointly manage the cost of TLS/QUIC session establishment, the amplified impact of a resolver outage, and the risk of a downgrade attack during the fallback to plaintext when the encrypted tunnel fails. Absorb latency with caching, connection reuse, and 0-RTT, but choose a policy of "reject on failure (strict mode)" rather than "plaintext on failure" according to the security grade.
Tension with regulation/censorship: Because encrypted DNS/ECH nullifies SNI-interception-based blocking, they may collide with national content policy. A professional engineer must present an adoption strategy that considers the balance between legitimate blocking (child protection, etc.) and user privacy, together with domestic law and privacy-protection requirements.
Standards-based discovery/upgrade framework: Since forcing the use of a hardcoded external resolver breaks split-horizon DNS and internal-domain resolution, it is preferable to use network-designated resolver discovery standards such as DDR (RFC 9462) and DNR (RFC 9463) to induce "automatic upgrade to the network-designated trusted resolver." This is a realistic compromise that reconciles privacy with operator control, and it works effectively only when the policies of the endpoint, browser, and network are designed consistently.
Migration/compatibility strategy: In the transition period where legacy Do53 and encrypted DNS coexist, align the fallback behavior on resolver non-support/firewall blocking with the security policy, and concurrently introduce complementary controls (EDR/termination logging) to fill the monitoring gap. An approach of applying it step by step starting from core assets/sensitive domains, measuring the effect, and expanding reduces risk more than a wholesale switch.
PQC/long-term outlook: Since encrypted DNS sessions depend on the key exchange of TLS 1.3/QUIC, a roadmap for transition to PQC hybrid key exchange against the quantum-computing threat must be established together. In addition, cultivating trusted domestic resolvers to mitigate concentration in a handful of global resolvers, phased introduction of an ODoH-style separated architecture, and policy-making on the principle of "encryption by default, control point with the organization" become mid-to-long-term tasks.
References
- RFC 9250, DNS over Dedicated QUIC Connections — https://www.rfc-editor.org/rfc/rfc9250
- RFC 8484, DNS Queries over HTTPS (DoH) — https://www.rfc-editor.org/rfc/rfc8484
- RFC 7858, DNS over TLS (DoT) — https://www.rfc-editor.org/rfc/rfc7858
- RFC 9230, Oblivious DNS over HTTPS (ODoH) — https://www.rfc-editor.org/rfc/rfc9230
In one line: Encrypted DNS (DoT·DoH·DoQ·ODoH) wraps the DNS queries that had remained in plaintext in TLS/QUIC to secure confidentiality on the transport segment; connection privacy is completed only when it is layered with DNSSEC (integrity), ECH (SNI concealment), and QNAME minimization, at the cost of leaving the governance challenges of "trust concentration on the resolver" and "loss of enterprise security visibility."