We blocked our own DNS resolver: when IDS alerts are not attacker evidence
InfraGuard Engineering Devlog, part 4 of 6. On October 25, we found Cloudflare's public resolver, 1.1.1.1, in our own blocking pipeline. Repeated IDS activity nominated sources; external reputation helped decide whethe
InfraGuard Engineering Devlog, part 4 of 6.
On October 25, we found Cloudflare's public resolver, 1.1.1.1, in our own blocking pipeline.
Repeated IDS activity nominated sources; external reputation helped decide whether to block them. Individual IP blocks could later be promoted to a /24 range when enough distinct addresses accumulated. Staging already blocked individual IPs; promotion expanded the scope.
The design failed when the counted evidence described our sensor rather than the remote host.
Unsolicited from whose perspective?
The signature was βSURICATA DNS Unsolicited response.β The record showed 1,572 events over seven days from just two sources, including 1.1.1.1.
A separate 30-day inspection found 1,328 records directed to our host's ephemeral UDP ports. These were responses to our queries, but the sensor no longer retained the request state.
Persistence nominated the resolver. Reputation did not rescue the decision: reports against a shared resolver made a misleading local signal appear corroborated.
Visibility failure is not attribution
Some alerts describe packets a remote endpoint sent. Others describe flow tracking, capture-thread distribution or lost context in the sensor.
Treating the latter as hostile-peer evidence can build the strongest-looking case against normal high-volume infrastructure. More legitimate traffic creates more opportunities for local visibility gaps.
Excluding the non-attributable signature category reduced candidates from 199 to 183. In that inspection, the removed 16 were CDN or cloud edges, including CloudFront, Amazon and Cloudflare infrastructure.
Fix the evidence class
We added a public-resolver allowlist, but an address list cannot fix a category error. The same failure could affect an unfamiliar resolver or CDN edge.
The durable change was a canonical code-level list of non-attributable signatures, shared by both candidate-selection queries. Whether a signature supports attribution is a property of the evidence, not merely an operator's tuning preference.
A November 12 change made allowlisting clean up matching existing block records too. βAllowed nowβ should not leave an earlier block silently effective.
Takeaway: Persistence strengthens evidence only after you establish what the evidence is about. Two systems agreeing on a shared false premise is not independent confirmation.
About this series: Adapted from InfraGuard engineering records. Measurements describe the recorded environment, not product-wide benchmarks. AI was used to draft and translate this article.
Disclosure: Solution Market develops and sells InfraGuard. This is our engineering devlog, not an independent review. InfraGuard is a self-hosted infrastructure operations and security platform. Project Β· Documentation Β· Solution Market
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.