France just watched a popular public resolver leave because a court asked it to block names. The irony writes itself. Quad9’s whole pitch is threat-protection at recursion; a French mandate treated that same lever as a content filter. Once you sell DNS as a policy plane, someone with a gavel will use it. If your cybersecurity program treats 9.9.9.9 as a control, you just watched the control grow legs.
Sysadmins have been quietly fine with that bargain for years. Point DHCP at a “secure” resolver. Tick the cyber security box. Sleep. The Quad9 exit is the invoice coming due for that laziness. A recursive lookup service can vanish from a jurisdiction, change its block list, or start NXDOMAINing names you still need. You don’t get a change ticket. You get a help desk queue full of people who “can’t get to the internet.”
Cybersecurity is not an anycast address you rent
Public recursive DNS was a convenience. Then vendors and nonprofits wrapped it in block lists and called it a product. That pitch is catnip for budget slides. One IP in DHCP, some threat intelligence at the edge of the lookup, and leadership hears “we filter malice.” What you actually bought is a third party with a different legal home, a different risk appetite, and zero obligation to keep serving your country.
Cloudflare has a dog in this fight. They run a public resolver and they would rather not inherit a French court list either. Read the post anyway. The engineering argument holds even when the messenger is obvious. Recursion is a shared cache in front of the whole internet. It does not know who asked, which office they sit in, or whether the name is a phishing kit, a newspaper, or a court-ordered takedown. Anycast makes that worse. The same address lands you on different metal depending on BGP. Geo-specific policy on a global resolver is a bedtime story.

Users already walked around this. They call it Chrome. Encrypted DNS to a resolver you don’t log, a travel VPN, a phone that never saw your GPO. If your defense in depth starts and ends at someone else’s recursive anycast, you never had depth. You had a default.
This is a bad look for anyone who sold NXDOMAIN as governance. Blocking malware names at recursion can still be useful noise. Treating that noise as a national filter, or as your only threat detection for “bad domains,” is how you confuse a hint with a control. Quad9 leaving France is what happens when those two stories collide in a courtroom.
Attackers already use DNS as a directory
While governments try to turn resolvers into cops, phishing crews are using DNS the way it actually works: as a map. SANS handlers pulled apart a credential-theft page that looked generic until they noticed it also wanted MX records. That’s not a curiosity. That’s mail-vendor recon sitting next to the password field.
If I know your MX, I know whether the next lure should wear a Microsoft logo, a Google one, or whatever on-prem relic you still expose. The kit is doing homework your “secure DNS” deployment never assigned to a human. Your users type a password. The page asks the internet how their mail is routed. You log neither, because the lookup never hits a resolver you own.
Sit with that for a second. One camp wants DNS to refuse names. The other camp wants DNS to describe your estate. Both camps are treating the lookup path as a security feature. Only one of them is currently getting value out of it, and it isn’t you.
Workstations have almost no honest reason to query MX for random domains. Mail servers do. Security gateways do. A laptop in accounting does not. If your recursive logs can’t show you that distinction, you are flying blind on a protocol every intrusion still touches. Incident response keeps rediscovering this the hard way: the first reliable breadcrumb was a lookup, and nobody retained it.
Put the blocking on gear you actually own
Renting a moral stance is easy. Operating a resolver is work. Do the work. You can keep using a public recursive service for guest Wi-Fi or for a lab. You cannot keep pretending it is a control that will still be there after the next injunction.
Immediate moves, then the boring ones you keep skipping:
- Inventory who resolves for whom. Endpoints, servers, jump boxes, guest networks, cloud egress, and phones. Write down DHCP, GPO, MDM, systemd-resolved, and any encrypted-DNS settings browsers shipped without asking you. If two teams give you two answers, believe the messier one.
- Stand up recursion you operate, even if it just forwards to an upstream today. Log queries. Retain them long enough that incident response can ask “what did this host look up?” without filing a vendor ticket. Sinkhole internally if you want a friendly landing page; don’t outsource the NXDOMAIN and call it done.
- Enforce resolver choice on managed fleets, then hunt the bypass. Encrypted DNS to a public IP is an egress policy problem. Your firewall should see unexpected resolvers the same way it sees unexpected SSH. That’s security hardening with a packet filter, not a prayer to anycast.
- Keep brute-force and login abuse on the actual login surface. If SSH, RDP, or a web portal is getting hammered, that’s rate limits, lockouts, and host controls in the ipban family (IPBan Pro if you already live on Windows auth). A public resolver hiding a C2 name you never logged is a consolation prize.
- Treat roaming as a different network. Laptops that leave the building will use whoever they want unless you give them a client that still sends query logs home. Test the failure: what breaks if 9.9.9.9 stops answering in your country tomorrow? France already ran that drill for you.
- Watch MX and similar metadata lookups from odd sources as a hunting signal, not as trivia. Pair that with your mail-flow telemetry. The goal is threat detection on the protocol you already run, not another dashboard that says “blocked 12,000 bad domains” and never names a host.
None of this requires a particular vendor. It requires admitting that recursion is infrastructure. You already know how to treat DHCP, NTP, and identity that way. DNS got a free pass because it failed open and the internet still loaded.
If the resolver can walk, it was never yours
Tell leadership the plain version. A court asked a public recursive operator to act like a national firewall. The operator left. Every “we use secure DNS” bullet on your last audit packet just became a footnote. The open internet crowd will talk about precedent, and they’re right to. Your problem is smaller and ruder: you cannot page a nonprofit in another country when the SOC needs a query log.
Policy people love DNS blocking because it looks cheap. One list, many users, no client agent. Bypass is a browser toggle. Overblock is a headline. Underblock is silent. That’s the whole product. If your board wants content control, put it on systems with identity, authentication, and an appeals path. If your board wants malware containment, put it on resolvers, proxies, and endpoints you can rebuild at 2 a.m.
Quad9’s withdrawal is a canary if you still think public recursion is a civic utility with an SLA. It is a volunteer with a block list and a lawyer. Volunteers get to leave. Lawyers get to pick the jurisdiction. You get to own the leftover risk.
Run DNS like it can be used against you. Because this week it was, twice: once as a cop job a resolver refused, and once as a directory a phishing page happily queried. Boring lookups, owned by you, logged on purpose. That’s the whole strategy. Everything else is a rental with a press release.
Sources
Take Control of Your Server Security
Don't let brute-force attacks slow you down. Try IPBan Pro risk-free for 30 days.
Secure. Automated. Lightweight.
Take Control of Your Server Security
Don't let brute-force attacks slow you down. Try IPBan Pro risk-free for 30 days.
Secure. Automated. Lightweight.
