A marketing team launches a campaign site on a cloud host in March. The campaign ends in May, somebody tidies up and deletes the app service, and the CNAME on the company domain stays exactly where it was. Nothing breaks. Nothing alerts. The name now points at a backend that anyone can claim, and the exposure dates from the deletion, not from the day an attacker finally parks content there. Continuous external attack surface monitoring is supposed to catch precisely this. In practice, plenty of programmes would not, and the reasons are rarely the ones the sales deck prepares you for.
What follows is six specific traps: the mechanism behind each, and what to change. They are the things that break in week two and week three, after the first excited look at the dashboard.
What “continuous”actually means in external attack surface monitoring
Continuous is not one clock. It is at least three, and vendors routinely quote the fastest one.
The first clock is discovery: how often the platform expands outward from your seed domains to find names and addresses it has not seen before. The second is service re-checking: how often it re-resolves DNS, re-probes ports and re-reads banners on assets it already knows about. The third is change alerting: how quickly a difference between the last observation and this one turns into something that lands in front of a human. A product can genuinely watch certificate transparency logs hourly while re-probing ports every few days. Both statements are true. Only one of them is on the marketing page.
That gap matters because the two clocks catch different things. Certificate transparency, the public append-only log of issued TLS certificates, surfaces a new hostname within minutes of a certificate being issued for it. A new open port on an existing host appears in no log at all. The only way to see it is to look again. If re-probing runs weekly, a management interface opened on Monday morning can sit unseen until the following Monday, regardless of what the headline cadence says.
Most of the raw signal comes from the same three places: certificate transparency logs, DNS (forward, reverse, zone transfers where they are foolishly allowed, and passive DNS datasets), and registration data from WHOIS and RDAP. Everything else is enrichment. Our write-up on how external discovery expands from a seed domain walks through the mechanics in more detail.
Ask for freshness per data source, in writing. One number is not an answer.
Trap 1: treating discovery output as an inventory
Discovery returns candidates. It does not return a register of what your organisation owns, and the difference is where most programmes go wrong in the first fortnight.
From the outside, several things look like you. A wildcard certificate covering a shared platform. A partner-hosted customer portal on a hostname under your domain. A marketing automation vendor’s CNAME. The infrastructure of a company you acquired four years ago, still running under its old brand on a former supplier’s hosting. An IP range on a CDN that thousands of organisations share. Attribution logic has to guess, and sometimes it guesses wrong in both directions.
The two failure modes are not symmetrical in how they feel. Claiming an asset you do not own produces noise, an awkward email to a supplier, and eventually a dismissal. Quietly dropping an asset you do own produces nothing at all, which is why it is the dangerous one. An inventory that silently shrinks looks like progress.
Two practical requirements. First, every claimed asset should carry its evidence trail: this name appeared in a certificate issued for your registered domain, this address is inside a range registered to your organisation, this host resolves from a record in a zone you control. If you cannot see why the platform thinks something is yours, you cannot adjudicate it. Second, build an explicit unconfirmed state into the inventory rather than forcing everything into owned or not-owned. “We are not sure, and here is who is checking”is a legitimate status. Binary classification is how real assets get deleted.
Trap 2: alerting on state instead of change
A daily email listing every exposed service in your estate is not monitoring. It is a report, and after about ten working days people stop opening it. The content is almost identical each morning, the one genuinely new line is buried in position forty-seven, and the filter rule that sends it to a folder gets written on a Thursday afternoon.
Change is the unit that matters. A host that did not exist yesterday. A port that was closed and is now open. A certificate issued for a name you did not request. A server banner that moved from one version to another. A DNS record that disappeared. Everything else is background.
Thresholds should differ by tier, because the estate does. A new open port on the payments gateway and a new open port on a dormant campaign microsite are not the same event and should not share a queue. Tiering the estate before you turn the taps on is the single cheapest thing you can do to reduce alert fatigue for the security team, and it pairs naturally with setting different scan frequencies by asset tier.
Route the quiet signals separately too. Confirmations, closures and “no change since last check”updates belong in a dashboard or a weekly digest, not in the channel that is meant to mean something is wrong. Mixing them is how a channel loses its meaning.
Trap 3: mistaking “likely affected”for “definitely affected”
Passive external observation infers. It reads a response banner, a set of headers, a TLS fingerprint, the shape of an error page, and concludes that a host is probably running a given version of a given product. It has not tested the flaw.
Three things break that inference regularly. Back-ported patches are the commonest by some distance: long-term-support distributions fix the vulnerability and keep the original version string, so a host that reports an old web server version may be fully patched. Reverse proxies and load balancers report their own identity rather than the origin’s. And some administrators set deliberately misleading banners, which is security theatre, but it still poisons your feed.
The practical consequence is that a passive finding needs an agreed escalation route into an active or authenticated check before anyone raises a change request on it. Decide now who signs that off, what the testing window is, and which systems are out of scope. Without that route, passive findings split into two bad outcomes: ignored, or converted into an emergency change that patches something already patched. We have set out the distinction between likely and definitely in passive versus active vulnerability detection at greater length.
External monitoring sits beside vulnerability scanning and internal tooling. It does not replace either, and anyone selling it as a replacement has told you something about their product.
Trap 4: watching only for CVEs
CVE matching is the easiest thing to put on a dashboard and frequently the least useful finding in the list. Real external compromise more often starts with something that has no CVE at all.
The wider class is mundane: an admin interface published to the internet because a firewall rule was written for a migration and never removed, a database or management port open to 0.0.0.0/0, a default credentials login page on an appliance, a directory listing exposing backups, an object storage bucket set to public during testing. None of these is exotic. They are the ordinary sediment of an organisation that has moved quickly, and they are directly exploitable without a single exploit kit.
Certificate hygiene deserves its own monitoring stream rather than a footnote. Expiry dates, weak or incomplete chains, wildcard certificates sprawling across business units, and certificates issued for names nobody internally authorised. That last one is worth a standing alert; certificate transparency exists partly so that mis-issuance becomes visible.
Then dangling DNS. A CNAME outlives the cloud resource it pointed at, the backend name becomes free, and whoever registers it next serves content on your domain. The window opens at deletion, not at hijack, and a monitoring setup that only reports once attacker content appears has missed the point of being continuous. Flag the record the day the backend stops answering. At that stage the fix is a DNS change and costs nothing, which is explained further in our piece on how subdomain takeover follows a dangling record.
Trap 5: no owner, no route, no closure
Here is a scenario that kills more monitoring programmes than any technical limitation. An acquisition’s seed domain expands to several hundred hostnames. A portion sit on a former supplier’s hosting. Nobody internally can say who owns them, nobody has credentials, and the supplier relationship ended with the integration project. The findings are real. They are also unactionable, so they age. Twelve months later, the backlog is the argument against renewal.
Map asset classes to teams before switching alerting on, not after. Marketing sites to the digital team. Partner-hosted portals to the relationship owner with a named contact at the partner. Cloud estates to whoever holds the account. Inherited infrastructure to a specific person with a mandate to decommission, because “nobody”is not an owner.
Routing matters as much as ownership. A finding that lands in a tool people already work in, the ticket queue, the SIEM, the team channel, gets handled. A finding that lives only in a vendor portal somebody logs into monthly does not. And close the loop properly: re-verify after remediation, record the fix against the asset, and let the asset record show that this host had this problem and it was resolved on this date. Remediation you cannot evidence is remediation you will be asked about again next quarter.
Trap 6: measuring volume instead of exposure time
Asset counts and finding counts rise when discovery improves. A maturing programme so reports worsening numbers, which is an uncomfortable conversation to have with a board that has been handed a chart going the wrong way. The metric is the problem, not the programme.
Two figures do the real work. Time to detect: the gap between an asset or exposure appearing on the internet and your platform surfacing it. Time to close: the gap between detection and verified remediation. Those two show whether continuous monitoring is achieving anything the previous quarterly scan was not. Everything else is context.
Track recurrence as well. The same misconfiguration reappearing on new hosts is not carelessness, it is a template, a pipeline or a default in a deployment module. Fix it there and the finding stops coming back.
A monthly set a board will actually read: median time to detect, median time to close by tier, number of assets in unconfirmed attribution, open findings older than thirty days with named owners, and recurring issue types. All of it comes out of data the monitoring already holds.
Getting through the first quarter
Spend the first thirty days deliberately. Confirm attribution and clear the unconfirmed pile down to something honest. Tier the estate. Tune thresholds against a fortnight of observed data rather than a guess. Only then open the alert taps to the people who will be woken by them.
Questions worth putting to a vendor: freshness per data source rather than a single cadence, what evidence is stored behind each attribution decision, whether active testing is available and how it is scoped, and what happens to the asset history when you dispute a claim. Our list of questions to ask an EASM vendor before signing covers the commercial side.
Two myths are worth correcting internally before they circulate. External monitoring in its default configuration observes; it does not exploit, and framing it as hacking your own company will cost you goodwill in change advisory. And being monitored does not make you more of a target, because the certificate transparency logs and DNS data it reads are public, and attackers were already reading them. The NCSC’s asset management guidance makes the same underlying point: you cannot protect what you have not accounted for. The certificate transparency project is worth a look if you want to understand why so much external discovery is possible at all.
The Short Version
- Continuous runs on three clocks: discovery, service re-checking and change alerting. Ask for freshness per source.
- Discovery gives candidates. Keep an evidence trail and an unconfirmed state, or you will delete real assets.
- Alert on change, not on state, and tier the estate so critical systems do not queue behind microsites.
- Passive findings say likely, not definitely. Agree the escalation route to an active check in advance.
- Misconfigurations, certificates and dangling DNS records matter more than CVE counts. Flag the record at deletion, not at takeover.
- No owner means no closure, and the backlog becomes the reason the tool is dropped.
- Measure time to detect and time to close. Counting assets rewards poor discovery.
Frequently Asked Questions
What does continuous external attack surface monitoring actually monitor?
It monitors the internet-facing footprint attributable to your organisation: hostnames, IP addresses, open ports and services, TLS certificates, DNS records and the software versions those services advertise. The useful products also track configuration issues such as exposed admin interfaces, public storage and records pointing at decommissioned backends. All of it is observed from outside, with no agent and no credentials.
How is continuous monitoring different from a point-in-time external scan?
A point-in-time scan is accurate at the instant it was taken and begins drifting immediately, because developers deploy, certificates are issued and firewall rules change in between. Continuous monitoring re-observes on a cycle and alerts on the difference between observations. The value is in the change detection, not in the raw scanning, which is why alerting design matters more than scan speed.
Does continuous external attack surface monitoring exploit or attack our systems?
Standard monitoring observes rather than exploits: it reads public logs and DNS data, resolves names, and makes ordinary connections to see which services answer. It does not attempt to confirm a vulnerability by triggering it. Active or authenticated testing is a separate, scoped activity that should be commissioned deliberately and signed off, not left running by default.
Why does monitoring flag assets we do not think we own?
Attribution works by inference from certificates, DNS and registration data, so shared CDN infrastructure, wildcard certificates, partner-hosted portals and an acquisition’s legacy estate can all look like you from the outside. Some of those flags are wrong. Some are assets you genuinely own and had forgotten, which is the harder conversation, and it is why the dispute process should preserve the evidence rather than simply deleting the record.
How quickly should a newly exposed asset show up in the inventory?
If a certificate is issued, certificate transparency makes the name visible within minutes, so same-day appearance is a reasonable expectation. A new open port on an existing host depends entirely on your re-probing interval, which is often measured in days. Set the expectation per data source, and judge continuous external attack surface monitoring on how fast it closes that second gap for your highest-tier assets. If you want the next layer of detail, read on about why a point-in-time scan is not enough.
