skip to content

Your inspection root can mint any name: what custody evidence shows a customer's auditor that no insider mints one silently, and what does it cost?

level: principalimportance: should knowfreq 30%

answer

  1. evidence means visible to someone else
  2. policy documents are not evidence
  3. the log must be held elsewhere
  4. constraints are checkable unaided
  5. detectable, not impossible

basics

~20 s

Controls whose failure a party outside your team would notice: a non-exportable hardware key, a witnessed generation ceremony, an issuance log held append-only by another owner, and name constraints the auditor can read for themselves. They buy detectability, not impossibility.

solid answer

~50 s

Be clear what evidence means to somebody who does not trust you: not a policy document, but controls whose failure would be visible to a party your team cannot silence. The credible set is a key generated inside hardware and non-exportable, with a witnessed ceremony record, so copying it is not a quiet act; dual control for any use outside the proxy's signing path; an issuance log of every certificate minted, with names and timestamps, shipped to storage another team or the customer retains and your operators cannot rewrite; name constraints in the root, which the auditor verifies directly by reading it on a device, the only item they need not take on faith; and reconciliation of issued names against decrypt policy by someone outside the team. Then say what you cannot prove: absence. The honest claim is that silent minting is detectable and attributable.

go deeper

for a junior

Know that operating a signing key means keeping a record of what it signed and storing the key so it cannot simply be copied, and that saying so in a document is not the same as showing it.

for a middle

Be able to describe hardware key storage, a witnessed generation ceremony, dual control and an issuance log, and explain what each one would make visible if somebody misused the key.

for a senior

Show you can distinguish controls an outsider can verify from ones they must take on trust, and be precise about the claim the evidence supports rather than overstating it.

for a principal

Own the trade: price each control, decide what you promise contractually versus what you can evidence, secure another team's commitment to hold the log, and get the detection owner to sign for any scope you choose not to decrypt.

## The question behind the question An auditor, or your own staff, is not asking whether you *intend* to behave. They are asking what would happen if you did not. Every answer that reduces to "our policy forbids it" fails, because the policy is written and enforced by the same team that holds the key. Useful evidence has one property: **its failure is visible to somebody who is not you.** ## The credible ladder **Hardware custody.** The key is generated inside a hardware security module and never exists outside it. Exporting is not an administrative action but a physically constrained one. This converts "a copy could exist" from an unfalsifiable worry into an event that would leave traces, and it is the single most expensive item on the list. **A witnessed ceremony.** Generation happens once, in a scripted session, with named witnesses from outside the operating team, and produces a signed transcript. It is the only moment when everyone agrees the key had no copies, and it is the anchor for every later claim. **Dual control.** No single administrator can invoke the key outside the proxy's normal path. This is genuinely inconvenient: it slows you down in the middle of the night when a certificate operation is on the critical path of a restoration, and that inconvenience is precisely what makes it evidence rather than a slogan. **An issuance record you do not own.** Every certificate the proxy mints is logged with its names and time, and shipped continuously to storage that a different team, or the customer, retains under append-only retention. If your CA operators can edit the record, the record proves nothing about them. This is an organisational control disguised as a technical one: it requires a team willing to hold your logs and to refuse you when you ask to change them. **Name constraints in the root.** This is the only item the auditor can check without trusting your account of it: they read the root as installed on a device and see for themselves which subtrees can never be minted for. Evidence an outsider can verify unaided is worth several times the evidence they must accept from you. **Reconciliation.** Somebody outside the CA team periodically compares the names in the issuance log against what the decrypt policy permits, and investigates the difference. Without this step the log is an archive, not a control. ## What you must refuse to claim You cannot prove that no certificate was ever minted silently. Logs are produced by a system whose operators are the population you are being asked about, and the honest formulation is: **minting outside policy is detectable and attributable, not impossible.** Auditors, and staff, place more weight on a team that draws that line clearly than on one that overstates. It also protects you contractually: a promise that "we cannot" becomes a breach the day someone demonstrates that you could. ## The cost, and the decision a lead owns | Control | What it costs | |---|---| | Hardware custody plus ceremony | Real money, procurement time, and a calendar with outside witnesses | | Dual control | Slower response when key operations sit on a restoration path | | Externally held issuance log | Another team's ongoing commitment, and their willingness to refuse you | | Name constraints | An exclusion list frozen until the next fleet-wide root distribution | | Reconciliation | A standing review owned by someone who does not report to the CA team | And the cheapest control of all is the one people skip: **reduce the scope of what you decrypt.** Categories you never intercept need no evidence, because nothing is minted for them. That reduction costs inspection coverage, so it is not the CA custodian's call alone; whoever owns the detection outcome has to sign for the blind spot they are accepting. There is a second audience with standing here. In a remote-first company, staff are also a party that does not trust the team by default, and in several jurisdictions interception of personal traffic on managed devices requires consultation, notice, and a published exclusion list before it is switched on. The evidence pack you build for a customer's auditor is largely the same pack that makes the staff conversation survivable, and building it once for both audiences is the efficient move. ## How to answer Name the controls, say which one the auditor can verify unaided, state plainly what cannot be proven, and then put a price on each so the person holding the budget can choose. A principal-level answer ends with a recommendation and its trade, not with a list.

  • Which item on that list can the auditor verify without trusting anything your team says?
    The name constraints in the root, read directly from the certificate as installed on a device. Everything else, the hardware attestation, the ceremony transcript, the issuance log, reaches them through your team or a system your team runs. That is why the constraint is worth disproportionately more as evidence than as a control.
  • Why insist the issuance log is retained by a team that can refuse you?
    Because the question is what happens if the CA operators misbehave, and a record they can rewrite says nothing about them. Append-only retention under a different owner turns the log into evidence about the operators rather than a convenience for them. The cost is a real commitment from that other team.
  • A contract draft says you cannot mint certificates for the customer's own domains. Do you sign it?
    Not as written, unless a name constraint excluding those domains is in the root and you accept that changing it needs a fleet-wide redistribution. Otherwise the truthful wording is that such issuance is out of policy, logged and reconciled, so it would be detected and attributable. Do not promise an impossibility you can only make unlikely.
  • What is the cheapest way to shrink the evidence burden altogether?
    Decrypt less. Categories you never intercept require no minting, no exclusion argument and no evidence. The trade is inspection coverage, so the owner of the detection outcome has to accept the blind spot explicitly rather than the CA custodian deciding alone.

saying these in an interview costs you the question

  • Offers a key-management policy document as evidence
  • Claims silent minting is impossible rather than detectable
  • Keeps the issuance log where the CA operators can rewrite it
  • Ignores staff as a party owed evidence
  • Never prices the controls for the budget holder

context