MITRE ATT&CK Enterprise · TA0001
Initial Access: how an attacker gets in
Initial AccessTA0001
Every breach has a first hour. This is what it looks like, what it costs, and the telemetry that catches it — the ATT&CK tactic that covers the front door.
For the owner
An intruder does not need to break in. Most of the time they walk in.
That is the finding that surprises business owners the most, and it is why security analysts talk about "initial access" as a tactic rather than an event. The way an attacker gets into an environment is usually something that was already there: an account that was never supposed to be used for logging in from outside, a password that was reused from a personal email, a web server that was left on the internet with a patch missed in March, or a supplier whose update server you trusted without checking.
The business consequence follows directly. A break-in that starts with a password costs nothing extra to begin with, so attackers try the cheapest door first, and they try it against everyone. That is why a single customer of a Canadian logistics firm saw credential stuffing against their customer portal three weeks before anyone noticed, and why a municipality found an old backup service account still enabled two years after it was retired.
Three questions are worth asking your team about initial access, and they matter more than any tool you might buy:
- Which accounts are allowed to sign in from outside, and who reviews that list? Valid accounts are the most-covered technique in this series and the one most often missed in practice, because a successful login looks exactly like a successful login.
- What is exposed to the internet right now, and when was it last patched? Every externally reachable service is a door, and the inventory of those doors is usually the weakest document in the building.
- What happens in the first ten minutes after a suspicious email? If the answer is "nothing, until someone notices", the door is open by default.
None of this is expensive to fix. Most of it is attention. The reason it is worth attention is that initial access is where the entire rest of the attack is decided: everything an intruder does after the first hour is a consequence of the first hour, and the organisations that detect early are the ones whose problem stays a small problem.
For the analyst
The tactic
MITRE ATT&CK files Initial Access under TA0001. It is the "why" behind a technique: the adversary is trying to get into your network. The matrix files 22 techniques under it, and the pinned SigmaHQ corpus carries 191 rules that detect techniques in this tactic.
Initial Access is deliberately narrow in ATT&CK terms. It is not "the attack"; it is the opening move. The techniques that follow it — execution, persistence, privilege escalation — are separate tactics, and an alert on Initial Access is the alert that arrives before any of them.
The techniques below are the ones the corpus actually detects, with the number of rules that name each:
| Technique | Name | Rules in the corpus |
|---|---|---|
T1078 | Valid Accounts | 51 |
T1190 | Exploit Public-Facing Application | 46 |
T1078.004 | Valid Accounts: Cloud Accounts | 37 |
T1133 | External Remote Services | 19 |
T1566.001 | Phishing: Spearphishing Attachment | 19 |
T1566 | Phishing | 11 |
T1078.002 | Valid Accounts: Domain Accounts | 5 |
T1078.003 | Valid Accounts: Local Accounts | 5 |
T1078.001 | Valid Accounts: Default Accounts | 3 |
T1189 | Drive-by Compromise | 3 |
T1195.002 | Supply Chain Compromise: Compromise Software Supply Chain | 3 |
T1200 | Hardware Additions | 3 |
T1566.002 | Phishing: Spearphishing Link | 3 |
T1195.001 | Supply Chain Compromise: Compromise Software Dependencies and Development Tools | 2 |
T1091 | Replication Through Removable Media | 1 |
T1195 | Supply Chain Compromise | 1 |
T1199 | Trusted Relationship | 1 |
The three ways in
Valid accounts (T1078 and its sub-techniques). The single largest block of rules under this tactic, and the one most likely to be dismissed as noise. ATT&CK splits it by the kind of account: default accounts (T1078.001), domain accounts (T1078.002), local accounts (T1078.003), and cloud accounts (T1078.004). The distinction matters because the telemetry differs. A domain account authenticating from a new address is a directory event; a cloud account authenticating from a new address is an audit log entry; a local account authenticating at all is a Windows event.
The detection problem is real: a successful authentication with valid credentials is indistinguishable from a legitimate one, except by context. What makes it detectable is the context — the source address, the time, the user agent, the destination. The DetectMonkey normalizer flattens these into the same columns for every source, so user_name, src_ip, dst_ip, event_type and severity are the fields a rule reasons over regardless of whether the event came from a Windows log, an auditd record or a cloud audit trail.
Exploit public-facing application (T1190). Exploiting a weakness in something you published to the internet. A web server, a file transfer service, a virtual private network concentrator, a remote management interface. The corpus has 191 rules under the tactic as a whole, and T1190 alone carries a large share of them, because this is the tactic with the most detectable signatures.
From the analyst's side, the useful signals are: a process on a server that spawns a child of a web process, a web server process making an outbound connection it has never made before, and file modification under a web root by a process that is not the web server.
Phishing (T1566, T1566.001, T1566.002). A message that carries a payload the recipient is meant to act on. The sub-techniques split by delivery: spearphishing attachment, spearphishing link. These are not the initial access itself — phishing is the delivery, and the attacker is still outside until someone runs the attachment. That gap is exactly what the detection has to cover, and it is why a rule that fires on "a file was downloaded and then executed" is worth more than one that fires on "an email arrived".
Where the telemetry comes from
For this tactic, the useful sources are narrower than for later ones. Endpoint telemetry sees the result of initial access (a process started by an unexpected parent), not the attempt itself. What sees the attempt is:
- Cloud identity audit logs. A sign-in from a new location, a new device, or a new user agent. This is where
T1078.004is detected, and it is the highest-value signal in the whole tactic, because it arrives at the moment of the intrusion rather than after it. - Firewall and network logs. The connection to an unexpected destination. A workstation in a manufacturing plant that has never spoken to a host in another country is a meaningful event on its own.
- Web server logs. The request that carried the exploit, and the response code it got. A spike of 404s followed by a 200 from the same source address is a scan that found something.
- Endpoint process creation. The first process of the intrusion.
The T1195 supply chain sub-techniques are the hardest of the lot: a compromised dependency or a trojanised update runs with the trust of the software that installed it, and there is no process event that looks wrong. Detection here leans on the update path itself — a signed binary whose signature does not match, or an update channel outside the expected one.
What the numbers do and do not tell you
A rule count is a coverage measure, not a risk measure. 191 rules map to this tactic because the corpus is large and this tactic is broad; it does not mean 191 independent ways to be attacked.
What it does mean: for most of the techniques in this table, there is a detection somebody has already written, tested against real corpora, and published. The scarce resource in initial access is not the detection logic. It is deciding which of those detections your environment should fire, and having someone look at the answer.
Detection engineering notes
If you are writing rules for this tactic yourself, three things decide whether they work:
- Anchor on the rare event, not the common one. "A sign-in succeeded" fires constantly and teaches people to ignore it. "A sign-in succeeded from an address never seen for this account" fires rarely and means something.
- Use the normalised columns, not the raw names. The corpus is not consistent about spelling —
CommandLineandcommand_line,Imageandprocess_name. The normalizer maps the collector's spelling onto one set of columns so a rule reads the same regardless of which log it came from. - Expect the first alert to be about the payload, not the door. The initial access event is often only visible in context; the executable that follows is what crosses a line. Rules that correlate the two — unusual origin plus unusual child process — are what make this tactic detectable rather than merely covered.
Where DetectMonkey fits
DetectMonkey does not stop initial access. Nothing on a shelf does, and any vendor promising otherwise is either blocking at the network edge, which is a different product, or misdescribing what it does.
What it does is watch for it. The collectors read the telemetry this tactic produces — endpoint process creation, cloud identity audit events, firewall connections — and the pinned corpus of 3164 Sigma rules evaluates them as the events arrive rather than after they are indexed, so an unusual sign-in and the process it started can be alerted on while the intruder is still in the building.
The three questions at the top of this post are answerable from that telemetry. Which accounts may sign in from outside is visible in the identity audit stream. What is exposed to the internet is visible in the connection logs and the host inventory. What happens in the first ten minutes is visible in the alert stream and the 72-hour window behind it.
The platform detects and alerts. It does not isolate, quarantine or block anything on your systems, and it does not change your configuration. Every alert carries the evidence that produced it, in both English and French, and the telemetry is processed and stored on private servers operated by Codein in New Brunswick, Canada — not in a public cloud.
Early access is 90 days at no charge, for up to 100 seats, with 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.