MADWeb @ NDSS 2026
NinjaDoH

NinjaDoH: A Censorship-Resistant Moving Target DoH Server
Using Hyperscalers and IPNS

1 University of Oklahoma
2 University of Texas at San Antonio
Workshop on Measurements, Attacks, and Defenses for the Web (MADWeb) 2026
NinjaDoH system diagram from the MADWeb 2026 paper.

NinjaDoH uses a decentralized control plane (IPNS/IPFS) and hyperscaler IP rotation to make DoH discovery a moving target—without breaking normal HTTPS traffic.

Abstract

We introduce NinjaDoH, a novel DNS over HTTPS (DoH) protocol that leverages the InterPlanetary Name System (IPNS), along with public cloud infrastructure, to create a censorship-resistant moving target DoH service. NinjaDoH is specifically designed to evade traditional censorship methods that involve blocking DoH servers by IP addresses or domains by continually altering the server’s network identifiers, significantly increasing the complexity of effectively censoring NinjaDoH traffic without disruption of other web traffic. We also present an analysis that quantifies the DNS query latency and financial costs of running our implementation of this protocol as a service. Further tests assess the ability of NinjaDoH to elude detection mechanisms, including both commercial firewall products and advanced machine learning-based detection systems. The results broadly support NinjaDoH’s efficacy as a robust, moving target DNS solution that can ensure continuous and secure internet access in environments with heavy DNS-based censorship.

Key contributions

  • Moving-target DoH design & implementation using hyperscaler IP agility and IPNS to distribute the latest server endpoint and query path.
  • Performance evaluation showing low-latency DNS resolution comparable to public DoH providers (and dramatically faster than DNS-over-Tor).
  • Censorship-resistance evaluation demonstrating evasion of common firewall blocklist techniques and reduced effectiveness of baseline ML detectors.
  • Cost analysis indicating the prototype deployment can be run for approximately $23.55/month (configuration-dependent).

System overview

NinjaDoH is built around a moving target defense: the server rotates public IP addresses on hyperscaler infrastructure, while a decentralized “control plane” (IPNS/IPFS) publishes the latest server IP address and a randomized DoH query path. The client resolves the IPNS record, updates a local DNS proxy, and routes DNS queries over HTTPS to the current server endpoint—preserving compatibility with existing browsers and applications.

Problem: Encryption hides the content, but not the traffic pattern. NinjaDoH focuses on defeating the patterns censors rely on: static lists, fingerprints, and long-lived flow signatures.

Control plane: IPNS/IPFS

Clients learn the latest server IP + query path via an IPNS name resolving to IPFS content (CID).

Data plane: HTTPS (DoH)

DNS queries blend into normal port 443 traffic while avoiding standard DoH fingerprints (e.g., fixed /dns-query paths).

Zero-downtime IP rotation

Overlapping IP assignments (“the ladder”) allow smooth handoff while clients refresh endpoints.

Fast certificate updates

A private CA enables rapid certificate issuance compatible with frequent IP rotations.

Results

Latency (mean)
12.68 ms
NinjaDoH's ping-adjusted performance is inline with public DoH providers (7.77 ms).
DNS over Tor (mean)
601.16 ms
NinjaDoH's performance is significantly more performant than DNS-over-Tor.
ML evasion (baseline)
Recall ≈ 0.506
NinjaDoH is effective against baseline ML detectors.
Adaptive adversary model
F1 = 0.578
Even if an adversary trains against NinjaDoH data, they acheive little gains (Best targeted model: precision 0.764, recall 0.635, F1 0.578).
Firewall evasion
4 / 5
Evades domain/IP blocking, application identification, and SNI blocking; strict allowlisting remains effective (but at significant cost to the censor).
Cost (prototype)
$23.55 / month
Costs to deploy are quite reasonable and scalable for small groups.

See the paper for full experimental methodology, confidence intervals, and threat model assumptions.

Slides

Tip: scroll horizontally (trackpad/swipe) or use the arrows to navigate. The download button above opens the full PDF (zoom/search).

BibTeX

@inproceedings{Seidenberger2026NinjaDoH,
  title        = {NinjaDoH: A Censorship-Resistant Moving Target DoH Server Using Hyperscalers and IPNS},
  author       = {Seidenberger, Scott and Beret, Marc and Wijewickrama, Raveen and Jadliwala, Murtuza and Maiti, Anindya},
  booktitle    = {Workshop on Measurements, Attacks, and Defenses for the Web (MADWeb)},
  year         = {2026},
  address      = {San Diego, CA, USA},
  month        = {Feb},
  doi          = {10.14722/madweb.2026.23006},
  url          = {https://dx.doi.org/10.14722/madweb.2026.23006}
}