ATT&CK guide
Ransomware, honestly: what detection does and does not buy you
No tool stops ransomware. What shortens an attack is finding it early — and that is a detection problem, a recovery problem and an operations problem, in that order.
For the owner
Somebody will tell you their product stops ransomware. Ask them: stop it how, and at what point? No monitoring product stops it. Prevention comes from a different family of tools — patching, multi-factor sign-in, offline backups, restricting what a normal account can reach — and those are worth doing regardless.
What monitoring buys you is time, and time is the whole game.
A ransomware attack is not one event. It is an intrusion that lasts hours or days, and the encryption is the last few minutes. Find out during the entry and you are dealing with one machine. Find out when the ransom note appears and you are dealing with the estate — and your backups may be gone. That difference is dwell time, and dwell time is decided at the front door, not at the encryption.
The cost of getting this wrong is not abstract: the ransom, or the rebuild, or both. Weeks of manual processing, a late payroll, and an obligation to notify people whose data was copied.
Three questions for your own team:
- When did anyone last watch a restore actually happen — not a backup succeed, a restore?
- If your backups were deleted tomorrow morning, where does the business start instead of resume?
- Who do you call at two in the morning, and what do they do in the first thirty minutes?
The third question decides everything else. This guide is arranged around answering it early enough to matter.
For the analyst
First, the distinction that gets argued about
Prevention is the set of controls that make the attack fail: the exposed service is patched (T1190), the account cannot be reused (T1078.004), the backup is not reachable from the identity that would be stolen, and the file server is not writable from a workstation.
Detection is knowing the attack is happening while it is still reversible. It stops no byte from being encrypted. It is the difference between discovering the intrusion on the way in and discovering it in the ransom note — and for a business of any size that is worth more than any single control.
Both are needed. There is also a third thing often confused with detection: recoverability. A perfectly detected attack with no offline, tested backup is still a catastrophe; one with an intact backup is an expensive Tuesday.
Why early access is the single biggest lever
Dwell time — the hours or days between the intruder arriving and the encryption starting — determines the blast radius. Short dwell time means one account on one machine and an answer of "reset the password, rebuild the laptop". Long dwell time means the intruder has already been through the file server, the backup system and the directory service, and the answer is "rebuild everything".
The lever is therefore not a better ransomware rule. It is entry detection: the exposed service (T1190), the valid account used from somewhere new (T1078.004), the credential dump (T1003.001), the first lateral move over SMB (T1021.002). Those land days before the impact, which is why they are worth the tuning effort.
The timeline of a ransomware night
The shape of the attack as the telemetry sees it.
- Entry. An internet-facing service is exploited (
T1190) or a cloud account is taken over (T1078.004). The first sign is a process started on a server by something that is not the web server, or a sign-in from an address the account has never used. - Credential access. Memory of the local credential store is dumped (
T1003.001), the directory database exported (T1003.003), the security account manager read (T1003.002). This is the pivot from one machine to the estate. - Lateral movement. Administrative shares over SMB (
T1021.002), remote desktop software (T1219.002), and PowerShell (T1059.001) used to run the next stage remotely. The file server is usually reached before anything is encrypted on it. - Staging. The ransomware binary is pulled in (
T1105), often obfuscated (T1027). Persistence is arranged: a Windows service (T1543.003), a scheduled task (T1053.005), a registry run key (T1547.001). - Preparation. The phase most organisations never see. Shadow copies are deleted (
T1490), backup catalogues emptied, recovery disabled, event logging tampered with (T1685.001), services stopped (T1489), administrator accounts removed or locked (T1531). - Impact. Files are encrypted (
T1486) or destroyed (T1485). Some operators compress an archive first and exfiltrate it before the encryption, so the copy may already be gone by the time you read the note.
What the telemetry shows, concretely
The signals that matter are in the normalised columns, and none says "ransomware detected":
- Mass file modification. A burst of file rename or file-write events on a host that normally writes a handful of files an hour. The extension changes, and so does the content's entropy.
- Recovery inhibition. A process calling the shadow copy API, a
wbadmincommand, abcdeditinvocation, a registry write disabling system restore. Rare enough to alert on with very little noise. - Service stops. A backup service, a database service or an antivirus service stopped by a process that is not the service manager.
- A new privileged identity. A domain account created, a user added to an administrators group, an account's roles changed — read from the identity audit trail, not the endpoint.
- Archive utility usage.
7z,rarorziprun against a directory tree that is not a build output, by an account that does not compress things. - Credential access.
lsassaccess by a process that is not the antivirus, antds.ditread,procdump, or a mimikatz-style signature.
The reason the last one matters is arithmetic: the encrypted file is a consequence, the dumped credential is the cause. An organisation that sees the encryption and the backup deletion and nothing else has already lost the ability to choose its response.
Why 72 hours of queryable history changes the outcome
This is easy to miss. When a machine is re-imaged after a ransomware event, the endpoint's evidence goes with it: the process tree, the command lines, the file operations. What survives is whatever was retained somewhere else.
The questions that determine whether you restore or rebuild are all historical:
- Which account started it? Was it an identity that lives in the cloud?
- Which host was first, and what reached it?
- Was the backup system touched before the encryption? If the backups were encrypted with a stolen key rather than deleted, restoring them restores the problem.
- Did anything leave the network before the encryption? If so, this is a notification obligation, not only an outage.
Seventy-two hours of continuously queryable telemetry covers the window in which most real intrusions begin, which makes it an operational requirement rather than a nicety. The longer archive behind it answers the other question — "has this host been affected before?" — which matters most for repeat incidents.
What a business should ask its provider
- Which of my systems would you actually be watching on the day of an incident? Not every asset has a collector, and the gap is where an intruder hides.
- When did you last run a restore, end to end, and how long did it take? A successful backup is not a restore.
- If I called at 3 a.m. and said "we think it is encrypting now", what happens in the first thirty minutes, and who is in that room?
- Which accounts can delete, alter or encrypt the backups — and would they also be stolen in a compromise?
- Do you receive the cloud identity logs — sign-ins, administrator actions, mail rules — or only endpoint telemetry? The entry point is often the identity, and without it the timeline starts wrong.
Where analysts get it wrong
The most common error is tuning towards the encryption alert. It is the loudest, it has the obvious business meaning, and it is the least useful, because it fires when the choice has already been made. The alerts worth protecting are the boring ones four hours earlier.
The second is assuming "ransomware" means "encrypted files". Some operators steal first and threaten later, and some never encrypt at all: the impact phase never happens, the extortion arrives by email, and the only detection you get is a phishing alert on the morning of the demand. The exfiltration was the incident.
The third is the false negative that costs the most: encryption performed on the file server, by the intruder, over the network. There are no endpoint file events, no rename storm on a workstation, and the impact telemetry sits on a machine your monitoring may not cover.
The fourth is a detection that fires on the right thing and gets dismissed. "Shadow copy deletion" is dismissible — the backup software does it, the storage team does it, and the alert has been seen a hundred times. The correlation is what saves it: the deletion that follows a credential dump and precedes a service stop is not the same event as the one in the noise.
Finally, the quiet one. Ransomware campaigns are now frequently paired with data theft, and the extortion message says so. A response plan that assumes the data is still on your systems will miss a notification deadline.
Where DetectMonkey fits
The honest version: the platform's job is to turn the first hours of an intrusion into an alert with evidence attached, and keep it queryable afterwards.
DetectMonkey collects the endpoint, identity and network telemetry the ransomware timeline is made of — process creation, file operations, service and registry changes, cloud identity events — and evaluates the pinned corpus of 3164 rules against them as the events arrive, not after they are indexed. That matters here more than in most tactics: an alert that arrives before the encryption rather than after is the difference between an incident and a rebuild.
It detects and alerts. It does not isolate, quarantine, block or remediate anything on your systems, and changes no configuration of yours. What it provides is the timeline an analyst needs to answer the questions above — which account, which host, how long ago, what else was touched — plus 72 hours of queryable history.
The telemetry is processed and stored on private servers operated by Codein in New Brunswick, Canada, and analysts 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.