MITRE ATT&CK Enterprise · TA0042
Resource Development: where the tools come from
Resource DevelopmentTA0042
Before the break-in, there is a shopping trip. This is what attackers buy, what it costs, and the telemetry that catches it.
For the owner
Most companies picture an attack as a break-in. It is usually closer to a delivery.
Before anyone touches your systems, an attacker has already collected what they need: a tool that does the job, somewhere to run it from, a codebase to adjust, an account that belongs to somebody else, a certificate so their traffic looks ordinary. None of this happens on your network. It happens on someone else's, and it is the reason the later steps look so legitimate.
What it looks like from the outside is that everything is slightly too correct. A message arrives from a supplier's genuine mailbox. An update installs from a domain that reads like the vendor's. A sign-in comes from a service you have already told your staff to trust. A warning about a certificate appears for a site that is not actually your site. The tell is not that something looks wrong. The tell is that nothing does.
You rarely pay this phase directly — somebody else buys the tooling, and often it is one of your own customers whose systems were used. But it determines what happens next: how credible the message is, which of your credentials the attacker already holds, and whether the tool they deploy is one your systems are supposed to refuse.
Three questions are worth asking your own team:
- Which third parties hold our credentials or our data, and would we notice if one of them were taken over? The answer is usually a list nobody has reviewed since onboarding.
- When a software update arrives from a vendor, what makes it distinguishable from a convincing copy? Most organisations have never asked, and most could not answer tomorrow.
- Which of our accounts, stolen today, would give somebody else a working foothold inside our business? Passwords and API keys are the same thing with different labels.
None of this is expensive to fix. Most of it is an inventory and a decision. It matters because the preparation phase is where the adversary's cost is lowest and where your own visibility is at its worst.
For the analyst
The tactic
MITRE ATT&CK files Resource Development under TA0042. It covers what the adversary acquires before they touch the target: the tooling, the infrastructure, the accounts and the malware itself. The matrix files 51 techniques under it, and the pinned SigmaHQ corpus carries 30 rules that detect techniques in this tactic.
The first question an analyst should ask about this tactic is where the activity physically happens. Most of it does not happen in your estate. A tool is bought on a forum and hosted on infrastructure you will never see. Malware is developed on an adversary's own build server. An account is purchased through a marketplace. That is why the corpus count is 30 rules against 51 techniques: the rules exist where the capability's shadow falls on something you do own — the moment it is downloaded, executed, staged, or used to send something from an account that should have been yours.
What the corpus detects
| Technique | Name | Rules in the corpus |
|---|---|---|
T1588.002 | Obtain Capabilities: Tool | 9 |
T1587.001 | Develop Capabilities: Malware | 7 |
T1584 | Compromise Infrastructure | 3 |
T1587 | Develop Capabilities | 3 |
T1586 | Compromise Accounts | 2 |
T1588 | Obtain Capabilities | 2 |
T1608 | Stage Capabilities | 2 |
T1586.003 | Compromise Accounts: Cloud Accounts | 1 |
T1588.001 | Obtain Capabilities: Malware | 1 |
T1608.003 | Stage Capabilities: Install Digital Certificate | 1 |
Five families sit under this tactic.
Obtain Capabilities (T1588, T1588.001, T1588.002). Acquiring the tooling rather than writing it. T1588.002 (Tool) carries the most rules in this table, which is the corpus's honest answer to a question most people get wrong: a lot of attacks are assembled, not invented. T1588.001 (Malware) covers acquiring ready-made malicious code rather than a tool.
Develop Capabilities (T1587, T1587.001). Writing or modifying the malware. T1587.001 (Malware) carries seven rules, and its detection surface is genuinely unusual: the build environment. A compiler invoked against a source file, a packer, an obfuscator, a generator that emits a payload from a template. None of these are malicious in isolation, which is exactly why this family is hard.
Compromise Infrastructure (T1584). Taking over someone else's server, domain or device to operate from. Three rules. The visibility is indirect and it comes from the use: the infrastructure appears in your logs as a source address, a destination domain, or the issuer of a certificate.
Compromise Accounts (T1586, T1586.003). Buying or stealing an account to use as infrastructure — a mailbox to send from, a cloud account to store in, a code repository to stage a payload. T1586.003 (Cloud Accounts) is the one with a rule, and it is the one that matters most in a cloud-heavy environment.
Stage Capabilities (T1608, T1608.003). Putting the capability somewhere the victim will fetch it: a public repository, a storage bucket, a paste site, a domain with a certificate installed. T1608.003 (Install Digital Certificate) is the sub-technique that makes an adversary's site look legitimate to a browser and to your mail filter.
Where the telemetry comes from
| Source type | What it contributes | The resource development signal |
|---|---|---|
windows_sysmon | Process creation, file writes | process_name and command_line matching an offensive tool; file_path in a staging directory |
linux_auditd | Process creation, file writes | The same on Linux hosts, including shell-outs to fetch tooling |
firewall_syslog | Perimeter connections | dst_ip and dst_port pointing at infrastructure you do not recognise |
keycloak | Identity events | New principals, new credentials, role grants on accounts that never administer anything |
m365_audit | Cloud identity events | Mailbox forwarding rules, new app consents, credential resets |
gworkspace | Cloud identity events | The Google workspace equivalents |
cloud | Provider-level events | Object listings and downloads from a bucket that is not yours |
The normalised columns a rule reasons over are the same in every case: tenant_id, host_id, event_ts, ingest_ts, source_type, service_name, event_type, severity, src_ip, dst_ip, src_port, dst_port, user_name, process_name, parent_process, command_line, file_path, message. The point of the mapping is that a rule written once reads the same whether the event came from a Windows endpoint or a cloud audit trail.
What makes an event rare, and what makes it wrong
The rare signals in this tactic are almost never "a known tool name appeared". They are combinations:
- An offensive tool's filename in
file_pathand that path being user-writable, and the parent process being a shell rather than an installer. - A file written to a temporary directory and executed within seconds of being written, then removed.
- A service, a scheduled task or a cloud role granted to a principal that has never held one.
- A first-seen destination domain, delivering a file, contacted by a host that does not normally fetch executables from the internet.
What makes them wrong, reliably, is legitimate double use. Remote administration software, credential-testing utilities, penetration-testing frameworks, log shippers and security scanners all live on real estates and all get flagged by name-based rules. The engineers who need them are the people who will mute the alert.
The distinction that survives contact with a real environment is context: who ran it, from where, under which account, and whether it did anything the tool is not normally for.
Detection engineering notes
- Anchor on the use, not the acquisition. A download is not a compromise. Plenty of legitimate software is downloaded. A rule that fires on the tool name alone will be muted by every team inside a week; a rule that fires on the tool executing from an unusual path, with an unusual parent, under an unusual account will not.
- Watch the staging path. The write-then-execute-then-delete sequence is much rarer than the download that preceded it, and it is a single host's behaviour, which makes it cheap to state accurately.
- Correlate account age with account use. For
T1586.003(Cloud Accounts), a brand-new principal that immediately starts sending mail or granting permissions is a different event from an eight-year-old service account suddenly used from a new country. Age is the discriminator that survives most environments. - Treat the build environment as a monitored surface. For
T1587.001, the interesting events are packers, obfuscators and generators — rare in a normal office, common in a build pipeline. Scope the rule to the hosts that should never be doing that. - Correlate the certificate with the payload.
T1608.003alone is weak — certificates are cheap. A newly-registered domain serving an executable to one of your hosts is not. - Allowlist sanctioned tooling by signed path and publisher, not by name alone, and time-box any red-team or assessment window rather than leaving a permanent hole.
Where analysts get this wrong
The first mistake is treating this tactic's rules as compromise. They are acquisition rules. Capability acquisition happens in legitimate environments every day, and a team that treats every one as an incident stops reading the queue.
The second mistake is assuming malware development is unobservable because it happens elsewhere. It happens elsewhere — but the artefacts often surface when the developed payload is built, tested or staged on infrastructure that your own systems touch.
The third, and the most expensive, is living off the land. If the adversary uses the interpreters, downloaders and installers that are already installed on every machine in your estate, there is no acquisition event at all. No rule for this tactic fires, because there was nothing to acquire. The intrusion still succeeds.
That is the false negative to keep in mind: an alert stream with nothing in it for resource development, no tool identified, and a successful intrusion anyway — because the "acquired" capability was a command that was already present.
Where DetectMonkey fits
The preparation phase is quiet by design, and no detection layer can make it loud. What a detection layer can do is close the gap between the moment a prepared capability is used and the moment anybody hears about it.
DetectMonkey does exactly that and nothing more. The pinned corpus of 3164 rules is evaluated as events arrive rather than after indexing, so the seconds between a staged tool executing and a mail rule being created are the seconds that matter. It does not isolate, quarantine or block anything on your systems, and it changes no configuration on them. Telemetry is processed and stored on private servers operated by Codein in New Brunswick, Canada, not in a public cloud, and the analysts working your alerts work in English and French.
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.