External Scan Frequency: Triggers Beat Calendars

External Scan Frequency: Triggers Beat Calendars

Most scan schedules were set once, by someone who has since left, and have never been re-argued. That is the real problem with external scan frequency: not that monthly is wrong or daily is right, but that the number was chosen to fit a change window and then defended for years as though it were a risk decision. Calendars are the weakest part of a cadence. They fire on dates; your estate changes on events.

What follows is a way to rebuild the cadence so it survives a change board and an auditor: a baseline per tier, a defined set of triggers that override it, a clean split between discovery and testing, and a measure that is not scan count.

Short version

  1. Run passive discovery continuously or daily across the whole seed-derived estate. It is cheap and it is the only thing that catches assets you did not create.
  2. Set active testing cadence by tier, agreed with named owners before it is written down.
  3. Define triggers that fire a scan regardless of the schedule: new certificate, DNS add or delete, new IP range, supplier change, vendor advisory.
  4. Treat deletions as triggers, not just creations.
  5. Report exposure time, not the number of scans you ran.

Why a calendar alone cannot carry your external scan frequency

An attacker does not wait for your window. A certificate appears in a public log, a name starts resolving, an advisory drops on a Friday afternoon, and the interesting work happens in the hours after. Your schedule, meanwhile, is a date in a spreadsheet. The gap between those two clocks is the whole argument.

A monthly cadence silently accepts a worst-case exposure window of a month. Quarterly accepts a quarter. Nobody writes that down in those terms, because written down it looks indefensible, but that is the deal the schedule makes on your behalf. If an exposed admin interface is introduced on day two of the cycle, you learn about it on day thirty.

The two frequencies people merge by mistake

Discovery frequency and testing frequency are different programmes with different costs, and almost every cadence document treats them as one number.

Discovery asks: what exists that I do not know about? Certificate transparency, passive DNS, RDAP and WHOIS, ASN and netblock attribution, public source scraping. None of it touches the target. It can run daily or continuously across the entire estate without asking anyone’s permission, and the marginal cost of running it more often is close to zero. We have written elsewhere about how discovery expands from a seed list into a candidate set, and the expansion only works if it runs often enough to catch things while they are new.

Testing asks: what is wrong with the things I know about? That is reachability, consent, sometimes stability. Authenticated checks, credentialed scans, aggressive plugin sets against a payment front end at 14:00 on a Tuesday: all of those need an owner who agreed in advance. Testing cadence is negotiated. Discovery cadence is not.

Merge the two and you get the worst of both. Either you throttle discovery to the rate your change board will tolerate for active testing, which caps coverage, or you push active testing to discovery speed and spend the next year apologising to application owners.

PCI DSS quarterly is a floor for one scope, not a programme

PCI DSS requires external vulnerability scans at least once every three months, performed by an Approved Scanning Vendor, for systems in the cardholder data environment scope. That requirement is real and you should meet it, but read what it covers. The PCI Security Standards Council sets that bar for a defined scope. In most organisations the in-scope estate is a narrow slice of the public footprint. It says nothing about the marketing subdomains, the supplier-hosted recruitment portal, the staging host named after the project that built it, or the DNS zone costing seven pounds a month that nobody has looked at since 2019.

Quarterly ASV scanning is a compliance floor for one scope. It is not a cadence design for the estate.

Set the baseline: cadence by asset tier and exposure type

Tiering is the part of this that is settled, and the tier-by-tier breakdown of external scan intervals covers the detail. In brief, four categories, and usually fewer than twenty entries in each for a mid-sized estate:

  • Revenue path and authentication. Checkout, login, SSO endpoints, APIs carrying customer data, VPN and remote access concentrators.
  • Supporting public services. Mail gateways, status pages, document portals, partner interfaces.
  • Marketing and campaign estate. Microsites, landing pages, agency-built things on supplier platforms.
  • Dormant and decommission candidates. Names that still resolve but should not, zones with no owner, hosts nobody will claim.

A workable starting pattern: continuous or daily passive discovery across everything derived from your seeds, active testing weekly on tier one, monthly on tier two, quarterly on tier three, and tier four handled by discovery plus a disposition decision rather than by testing at all.

The counter-intuitive part is tier four. Dormant and third-party-hosted assets have almost no business value and deserve more discovery attention than their value suggests, because they are where takeover lives. A record pointing at a deleted bucket or a closed SaaS account is not a vulnerability in the usual sense and a vulnerability scanner will often report it as healthy. The mechanics of that failure mode are covered in our write-up on dangling DNS records and subdomain takeover.

Agree the cadence with asset owners before it goes in the document. A schedule imposed on a team that was not asked gets suspended during the first freeze and never reinstated.

The change triggers that should override the schedule

Here is the shift. The baseline is the floor; triggers are what actually protects you.

New certificate issuance

A certificate appearing in a transparency log is frequently the earliest external signal that a new asset exists. Often before DNS resolves. Almost always before a change ticket is filed. Monitor the logs for your apex domains and for names in subject alternative names you did not expect, whether you query crt.sh, consume tiled logs directly, or use a tool that speaks the Static CT API. A new name you cannot attribute is a scan trigger and an attribution question on the same day.

DNS adds and, more importantly, deletes

New CNAME, new A record, a record pointing at a provider you stopped paying in March: all triggers. Deletions matter just as much and are the frequency blind spot nobody plans for. A monthly cadence will cheerfully keep reporting an asset as healthy weeks after the resource behind it was deleted at the provider, because the CNAME still resolves and the scanner still gets an answer from something. Creation events get attention. Deletion events get a trigger too, or you will find out from someone else.

Infrastructure and supplier change

A new ASN or IP range, a cloud region migration, a supplier switch, a decommission ticket closing. Each one changes what is reachable. The decommission ticket is worth wiring directly into the trigger list: the close of the ticket should fire a verification scan, because “we turned it off”and “it stopped answering from the internet”are not the same statement.

External events

A vendor advisory or an NCSC patch wave affecting perimeter technology lands on a Friday. At 09:00 on Monday someone asks which internet-facing hosts run the affected product and version. A cadence that last looked three weeks ago cannot answer that, and the honest reply, “we will know by Thursday”, is the one that ends up in the incident report. Advisory publication is a trigger, and the scan it fires is narrow: the technology, the port, the fingerprint.

Who gets to fire an out-of-band scan

Name them. Typically the exposure owner, the on-call security lead and the service owner for the asset in question. Give each trigger a pre-approved scope and a maximum intensity so the authorisation argument happens once, in advance, not at 23:00. The failure mode here is privilege creep: everyone can fire a scan, so everyone does, and within two months the triggered runs are noise and people stop reading them. Log every out-of-band scan with who fired it and why, and review that log quarterly. If a trigger produces nothing real for two quarters, retire it.

Passive, active, and the frequency trade-off between them

Passive signals run continuously at near-zero cost to the target. Active checks cost reachability, consent and occasionally stability. Match the method to the question: discover and attribute passively, confirm actively and less often.

Which brings up the thing high-frequency scanning does not fix. Reading a version banner ten times a week does not make the version more accurate. Back-ported vendor builds keep the old version string while carrying the patch. Fronted and virtual-hosted services answer for a CDN or a reverse proxy rather than the origin. A version string read weekly is no better evidence than the same string read monthly, because the problem is evidential, not temporal. The gap between passive inference and active confirmation is about method, not interval.

This matters for change boards. Asking for an emergency window on the strength of “the banner says 7.2″tends to fail. Asking on the strength of a safe, non-destructive check that demonstrates the vulnerable behaviour, with the request packet and the response recorded, tends to succeed. Decide in advance which confirmation checks are pre-authorised for which tiers, so the trigger and the evidence arrive together.

Measuring whether your cadence works: exposure time, not scan count

Scan count rewards the wrong behaviour. It is easy to produce and easier to pad. Four metrics are worth reporting instead:

  1. Time to inventory. From an asset first appearing externally (certificate timestamp, first passive DNS observation) to it sitting in the verified inventory with an owner.
  2. Time to detect. From the exposure being introduced to it being detected. Where you can reconstruct the introduction time from change records or log timestamps, do; where you cannot, say so rather than guessing.
  3. Time to close. From detection to the exposure no longer being reachable.
  4. Trigger hit rate. What proportion of real findings arrived from triggers versus scheduled runs. If triggers dominate heavily, your baseline is probably too slack for that tier. If scheduled runs dominate, your triggers are badly specified.

Record the cadence decision in the inventory itself, next to first seen, last seen, confidence, disposition and review date. The cadence is an attribute of the asset, not of the scanner, and when an auditor asks why this host is tested weekly and that one quarterly, the record answers without anyone searching their inbox.

UK regulatory pressure and where compliance frequencies run out

The Cyber Security and Resilience Bill was introduced to Parliament in late 2025 and, at the time of writing, is still progressing, so treat the detail as provisional and check the final text. The government’s published policy position proposed tighter incident notification clocks, with an initial notification well inside a day of becoming aware and a fuller report following. What those clocks test is detection speed, not scan speed. If an exposure sat undetected for five weeks, the clock did not start late out of kindness; it started when you became aware, and the five weeks before that is the part you will be asked about. Our note on what the Bill means for external exposure work goes further on scope.

The same logic applies to patch waves. The National Cyber Security Centre publishes advisories on a schedule that is not yours, and readiness means being able to answer the “which of ours run this”question within hours.

One last thing on auditors, because it surprises people. A documented, owned, evidenced cadence with a slower interval usually tests better than an undocumented continuous one. “We scan all the time”with no record of what was in scope, what the trigger list is, who approved it or when it was last reviewed is harder to evidence than “tier one weekly, triggers on certificate and DNS change, reviewed 15 September 2026 by the named owner, here is the log”.

Before you tighten external scan frequency, tighten attribution. A faster cadence across an estate you have only half attributed just produces the same blind spots more often. Start with the discovery and inventory material, particularly how forgotten assets surface in the first place, then come back and set the intervals.

Frequently Asked Questions

How often should external scans actually run?

Passive discovery should run daily or continuously across the whole estate, because it costs the target nothing. Active testing should follow tier: roughly weekly for revenue-path and authentication assets, monthly for supporting public services, quarterly for low-value marketing estate. Those intervals are the floor, overridden by triggers whenever the estate changes.

Is quarterly external scanning enough for PCI DSS compliance?

For the defined cardholder data environment scope, PCI DSS requires external scanning at least every three months by an Approved Scanning Vendor, plus rescans after significant change. That satisfies the requirement for that scope. It is not a cadence for the rest of your public estate, which typically includes the subdomains and supplier-hosted services where takeover and forgotten exposure actually occur.

What is the difference between discovery frequency and testing frequency?

Discovery asks what exists that you do not know about, using passive sources such as certificate transparency, passive DNS and RDAP. Testing asks what is wrong with assets you already know about, using active checks against live services. Discovery can run continuously without owner consent; testing needs agreed windows, which is why the two cadences should be documented separately.

Which events should trigger an unscheduled external scan?

New certificate issuance for a name you did not expect, any DNS record added or deleted, a new ASN or IP range, a migration or supplier switch, a decommission ticket closing, and a vendor advisory or NCSC patch wave affecting technology on your perimeter. Deletions deserve a trigger as much as creations, because a deleted backend behind a live CNAME is exactly what a scheduled scan misreports as healthy.

How do I prove my external scan frequency is defensible to an auditor?

Show the tier definitions, the cadence per tier with a named owner, the trigger list with who may fire an out-of-band scan, the log of triggered runs, and the review date. Pair that with exposure-time metrics rather than scan counts. A documented interval with evidence behind it stands up better than an undocumented claim of continuous coverage.