Canadian data residency · made, hosted, and operated in Canada

Early access — 90 days →

ATT&CK guide

The attack kill chain: what an intrusion actually looks like, hour by hour

Most people picture a breach as a moment. It is a sequence, and it has an order. Knowing the order is what tells you where you can still interrupt it.

Codein Cybersecurity November 22, 2026 5 min read

For the owner

A breach is not a moment. It is a sequence, and it happens in a fairly predictable order: someone gets in, they make the computer do something, they stay there, they look around, they move to the next machine, they take the data, and they make it hurt. That order is why "we would have caught it" is usually a false claim — a business catches the middle of the chain, or the end, and by then the damage is already done.

Two numbers matter more than any tool purchase, and neither of them is on a vendor's price sheet. The first is how long an intruder can stay on your network before anyone notices. For a small Canadian organization with good endpoint protection and no one watching the logs, that number is measured in weeks. The second is how long your data is reachable after the intruder leaves. The first is a detection problem; the second is a data problem, and it is the one that shows up in the newspaper.

Three things shorten the chain, and none of them require a large purchase:

  1. Someone is watching the telemetry, not just collecting it. Detection that nobody reviews is the same as no detection, except it produces confidence.
  2. The history is queryable while it matters. Most teams can look at what happened an hour ago; very few can look at what happened three days ago, which is the window that decides whether an incident is a bad afternoon or a rebuild.
  3. The accounts that can be used from outside are few and known. Every extra door is a place an intruder can start without touching anything clever.

The uncomfortable truth is that the chain is not broken by any one of these. It is broken when they are combined, and it is broken much earlier when it is broken early — which is the entire argument for treating the first tactic as the important one.

For the analyst

The chain, in order

ATT&CK does not draw a single "kill chain" diagram; it draws a matrix of tactics, and an intrusion moves through them in roughly this order:

StageTacticWhat the adversary is doing
1Initial AccessGetting in — a valid account, an exposed service, a phishing message
2ExecutionMaking a machine run what they sent
3PersistenceMaking the access survive a restart and a patch cycle
4Privilege EscalationGetting from an ordinary account to an administrator
5Defense Evasion / StealthHiding in plain sight so the tooling misses it
6Credential AccessTaking the keys to everything else
7DiscoveryLearning the shape of the environment
8Lateral MovementWalking to the next machine
9CollectionGathering what they came for
10Command and ControlKeeping the channel open to direct it
11ExfiltrationTaking the data out
12ImpactMaking it hurt

The ordering is not a rule, but it is a strong tendency, and it is a useful diagnostic. When you investigate an alert, ask which stage the adversary is at. A valid-account sign-in from an unfamiliar address, followed an hour later by a process on the same host with a parent it has never had, followed that evening by the same account authenticating to a second host — that is a chain, and each event only makes sense when you read the other two with it.

Where the chain actually breaks

Stage 1 is the cheapest to break and the most often missed. Valid accounts (T1078) dominate the corpus for a reason: the credential is the perimeter now. An intruder using a stolen credential generates no process event, no suspicious binary, no malware signature. The only evidence is a sign-in that differs from the account's own history.

Stage 3 is where you lose the ability to see. Persistence (T1112 for registry keys, and the whole family of scheduled tasks, services and startup entries) means the adversary is now a thing your systems expect. Every subsequent alert is judged against a baseline that the intruder helped write.

Stage 6 is the point of no return. Credential access (T1003 and the rest) converts one compromised account into a compromised environment. Before it, you can reset one password. After it, you need to reset everything, and you need to know what "everything" is.

Stage 11 is what the customer hears about. Exfiltration (T1567, T1048) is often the first time anyone outside the organization has any idea, and by then the notification obligations have started.

The telemetry that shows the chain

The single most useful thing an analyst can do is read one entity across several stages rather than one entity within one stage. The platform normalizes every source onto the same columns, so a pivot is a query, not a project:

  • user_name — the thread that ties stages 1, 4, 6 and 7 together
  • host_id — the thread that ties stages 2, 3, 5 and 8 together
  • process_name and parent_process — the thread that ties stage 2 to everything after it
  • src_ip and dst_ip — the thread that ties stage 8 to stage 10
  • event_ts and ingest_ts — the ordering, and the delay

The classic example: one account, three hosts, fourteen hours. A sign-in from an address never seen for that account. A process on host A whose parent is not the usual parent. The same account, from the same address, authenticating to host B an hour later — and host B had no activity for eight months. Read alone, three alerts. Read together, an intrusion.

That is also why the 72-hour window matters concretely rather than rhetorically. The third event arrives hours after the second, and the correlation that identifies the pattern usually needs all three. If your history starts at the moment the first alert fired, you have lost the thing that proves it was an intrusion.

What breaks the chain: detection, and its cost

Every stage above has detections. The corpus carries rules across all of them, and the volume is uneven in an instructive way — execution and persistence have the most rules because those are the most common ways to be noticed, and credential access and discovery have fewer because they are quieter and harder.

The practical cost of detection is not the rule. It is the alert, and it is the review. A chain of twelve stages where each stage fires three rules is thirty-six alerts, and if nobody reads them, the chain is unbroken. If somebody reads them, the chain breaks at the first one they understand.

That is the real argument for correlation over volume: a rule that fires on "an unusual parent process for this account, on a host where that account has not run before" is worth more than six rules that each fire on one stage. The former is the chain; the latter are the pieces.

The stages a small organization cannot cover alone

Two stages are worth being honest about:

Impact (T1486 for encryption, T1490 for recovery inhibition, T1489 for service stop) is not a stage you detect well in time. The signal arrives when it is happening, and the only control that matters is the one you set up beforehand — tested backups, immutability, and a rehearsed restore.

Exfiltration is the stage where detection is most valuable and most often absent, because it requires volume-based reasoning: this week's egress versus last week's, from this host versus its peers. That is a rule you can write, but only if the egress data is collected at all.

The honest summary

The chain is a sequence of small, individually unremarkable events. Detecting it is not about catching a dramatic moment; it is about noticing that event seven does not belong with events one and two. That capability is a function of collection coverage, history depth, and someone reading — in that order.

Where DetectMonkey fits

DetectMonkey is built for exactly this shape of problem: it collects the telemetry from every stage of the chain, normalizes it onto one set of columns, and evaluates the pinned corpus of 3164 Sigma rules against it as the events arrive — so the sign-in and the process it started are correlated while the intruder is still in the building, rather than found in a report next quarter.

The chain view is the product's reason to exist. The last 72 hours of telemetry stay queryable, which is the window that decides whether an incident is a bad afternoon or a rebuild, and the cold lake keeps 365 days for the cases where the question arrives late. Every alert carries the evidence that produced it, in English and French.

What it does not do is isolate, quarantine, block or remediate anything on your systems — it detects and alerts, and leaves the decision with your team. Telemetry is processed and stored on private servers operated by Codein in New Brunswick, Canada, never in a public cloud.

Early access is 90 days at no charge, for up to 100 seats, founder-led onboarding, and no card required.

See it on your own telemetry

Early access is 90 days at no charge, for up to 100 seats, founder-led onboarding, no card required.