skip to content

As a registrar's abuse engineer, do you suspend 30,000 pre-computed botnet domains?

level: principalimportance: nice to knowfreq 20%

answer

  1. the charter stops at names you sponsor
  2. most of the crop is registered by nobody
  3. generated strings collide with real businesses
  4. redelegation makes you the recipient
  5. acting alone just moves them one rung

basics

~20 s

Mostly no, and never on the list alone. A registrar can act only on names it sponsors, with evidence and legal cover. Most of a crop is unregistered or held elsewhere, and acting early just moves the population one rung down.

solid answer

~50 s

Start with the charter: a registrar controls the names it sponsors and nothing else — it cannot touch an infected machine, and has no authority over crops sitting in other registries under other law. So the list splits: a small number you sponsor and can act on, a majority nobody has registered where the only lever is a costly pre-emptive reservation, and a residue owned by someone else, some of it legitimate because generated strings collide with real words. For the actionable few you need evidence per name, a choice between suspending and redelegating, and legal sign-off, since a wrong suspension injures a paying registrant. Then the real question, timing: acting on one layer alone teaches the operator which layer you reached and walks the population onto the next, so an uncoordinated suspension can be worth less than waiting for a coordinated one.

go deeper

for a junior

Understand the boundary first: a registrar acts on names it sponsors and cannot reach an infected machine, so nothing it does is remediation.

for a middle

Be able to split a computed crop into what is sponsored, what is unregistered, what belongs to other registries and what collides with legitimate businesses, and say which of those is actionable.

for a senior

Show the operational choice: suspension against redelegation, the evidence needed per name, and why an uncoordinated action simply moves the population to the next compiled path.

for a principal

Own the negotiation — what your charter can honestly promise a caller who wants everything actioned today, who signs off contractual and data duties, and why holding for a coordinated window across every holder of a layer is the higher-value decision.

## The chair, and what it can actually reach An abuse engineer at a registrar or hosting provider sits on the other side of the fence from everyone who owns a victim estate. The levers are narrow and specific: suspend or hold a name the registrar sponsors, change its delegation, terminate a hosting account, refuse a registration. That is the entire toolkit. It does not include removing code from a machine, notifying the people who are infected, or even knowing who they are. Any answer that drifts into cleaning up victims has left the chair. This matters because it bounds the promise. Suspending a rendezvous name removes one pre-committed path. The implants continue to run, continue to wake on their trigger, and continue down the ladder to the next thing the author compiled in. ## Splitting the list A crop computed from a sample is not a list of things you can act on. It divides: - **Names you sponsor.** Actionable, and the smallest slice. Each needs evidence linking that specific registration to the family, not just membership of a computed list, because acting is a contractual act against a paying customer with a right to complain. - **Names nobody has registered.** The majority. There is nothing to suspend. The only lever is a pre-emptive reservation or block of the string, which costs money per name, must be renewed, and is a policy decision rather than an abuse one. - **Names registered elsewhere, or in registries you do not operate.** Someone else's decision entirely, under someone else's law, at someone else's pace. Country registries in particular have their own processes, evidence standards and languages. - **Collisions.** Generators emit pronounceable strings, and some will already be legitimate businesses. Every batch action needs an exclusion pass, and getting this wrong is the mistake that ends up in a lawsuit rather than a write-up. ## Suspend, or redelegate? Two very different acts, and the difference is not technical vanity. **Suspension** stops the name resolving. It is cheap, fast, reversible, and it puts the population beyond anyone's sight: the hosts try, fail, and move down the ladder without a trace anyone can hold. **Redelegation** points the name at nameservers you control, so the population keeps arriving — at you. That is materially heavier. You are now receiving connections from every infected machine, which means addresses belonging to identifiable people and organisations flowing into your systems: a data-protection question about lawful basis, retention and who may receive it, and a question about what you are obliged to do with knowledge of who is infected. It usually requires a court order or a formal arrangement with law enforcement, and an infrastructure commitment for as long as the population lives. It also, unlike suspension, is distinguishable to the implant: a name that answers with an address that does not behave like the operator can trip anti-takeover checks and push hosts down the ladder anyway. ## The timing problem, which is the actual judgment Everything above is procedure. The judgment is *when*, and it comes from the fact that the ladder is pre-committed. The operator cannot revise it, so each layer you take is one they cannot get back — but only if they cannot simply step down. Take one path in isolation and you have told the operator which layer you reached, converted a population you could still find into one you cannot, and left them the rest of the ladder. Coordinated action across every registrar and registry holding a layer, at one moment, is worth far more than any single suspension; that coordination is slow, involves parties with different legal exposure, and is frequently what a careful engineer is waiting for rather than being obstructive. Which leads to the honest position to hold in front of a caller who wants the list actioned today: name what suspension delivers (one path, this period), name what it does not (any change on any infected machine), name the cost of doing it uncoordinated (the operator learns, and the population walks to the next layer), and say what would change your answer — sponsorship of the names, evidence per name, legal cover, and a coordinated window with the other holders. ## What a good answer sounds like Refuse the framing that thirty thousand computed strings are thirty thousand actionable items. Split the list, name the evidence and contractual bar, distinguish suspension from redelegation on duties rather than mechanics, and finish on coordination and timing — because that is the part that decides whether the action was worth taking at all.

  • Why is pre-emptively blocking unregistered generated strings a policy decision rather than an abuse one?
    Because there is no registrant and no incident to act on — you are reserving inventory against a prediction. It costs per name, needs renewing every period, and permanently denies strings to legitimate applicants, so it belongs to whoever owns registry policy and budget rather than to a case queue.
  • The caller says every hour of delay leaves people infected. What is the honest answer?
    That suspension does not change anything on an infected machine, so the hour costs no cleanup. What it changes is which path the population uses next, and that is worth spending once, coordinated across every holder of a layer, rather than early on one. Say plainly what your action can and cannot deliver, and what would make you act now.
  • What makes redelegating a seized name heavier than suspending it?
    It makes you the destination for every infected host, so identifiable victim addresses arrive in your systems and you need lawful basis, retention limits and a decision on who may receive them. It normally needs a court order or a law-enforcement arrangement, plus infrastructure for the life of the population, where suspension is reversible and holds no personal data.

saying these in an interview costs you the question

  • Treats a computed crop as a list of actionable registrations
  • Assumes suspension removes the infection or helps victims directly
  • Skips the collision check with legitimate registrants
  • Ignores that acting early pushes the population down the ladder
  • Confuses redelegating a name with simply suspending it

context