Blog

Adversary Emulation: testing your defense against a real attacker

How adversary emulation works with MITRE ATT&CK: the emulation plan, technique-by-technique runs, purple teaming and what to ask a supplier.

Light and dark wooden chess pieces on a board in half light

A pentest answers whether an exploitable flaw exists. A Red Team answers whether an objective can be reached by any permitted path. Adversary emulation answers a third question, narrower and very useful to anyone running a SOC: if a known group attacked your company the way it usually attacks, would your controls see every step?

To answer it, emulation starts from a concrete adversary, turns its behavior into a script and runs that script against your environment, measuring what each control logged, alerted on or let through.

The vocabulary: MITRE ATT&CK

Emulation rests on MITRE ATT&CK, a public knowledge base of attacker behavior observed in real incidents. It has three levels:

  • Tactics: the goal of each phase, such as initial access, execution, persistence, privilege escalation, credential access, lateral movement, exfiltration and impact.
  • Techniques and sub-techniques: how the attacker meets each goal. T1566.001 is spearphishing with an attachment; T1059.001 is execution through PowerShell; T1003.001 is dumping credentials from the memory of the LSASS process; T1558.003 is Kerberoasting.
  • Procedures: the specific implementation a group used, with the tool, the command and the order in which it appeared.

ATT&CK also keeps pages on groups, with the techniques attributed to each and the public sources behind that attribution. The adversary’s profile is built from that material.

Using this vocabulary makes results comparable. When the report says T1021.002 (lateral movement through Windows admin shares) went through without an alert, the detection team knows exactly what to look for, and the next exercise can measure whether the gap closed.

How an exercise works

1. Choosing the adversary

It starts with a question: who attacks companies like yours? A bank, a retailer and a manufacturer face different groups. The choice draws on threat intelligence, public reports and ATT&CK’s group pages. For companies in Latin America, for example, ATT&CK documents groups with campaigns in the region, such as Blind Eagle (G0099), known for spearphishing and for widely available remote access trojans.

You can also emulate a behavior instead of a group: a ransomware operator’s typical flow, from initial access to encryption, covers techniques many groups share.

2. The emulation plan

The plan turns the profile into a script you can run. For each step it sets the technique, the procedure, the order, the prerequisites and the signal expected from each control: what the EDR should record, which SIEM rule should fire, which SOC use case should open a ticket.

There are public references for this. The Center for Threat-Informed Defense publishes a library of emulation plans for groups such as FIN6, FIN7, APT29 and Wizard Spider, with the sequence of techniques and supporting code. For atomic tests of a single technique, the Atomic Red Team project keeps small tests mapped to ATT&CK. Platforms such as Caldera, created by MITRE and now in the Apache Incubator, automate part of the execution. These help, but the plan has to fit your environment: the tools, operating systems and controls you actually have.

3. Controlled execution

Techniques run in the authorized windows and environments, with care not to destroy data or interrupt services. An impact technique such as T1486 (data encryption) can be emulated on test files created for the exercise. Every run is logged with time, machine, user and command, so what happened can later be matched with what the defense saw.

4. Technique-by-technique analysis

This is where the exercise pays off. Each technique lands in one of three categories:

  • Not seen: no control produced useful data.
  • Detected: there was telemetry (a log, an EDR event), but no alert.
  • Alerted: an alert reached the SOC, and the team could act.

The gap between the second and third categories is where many companies find their weaknesses. The data is in the SIEM, but no rule turns it into an alert, or the rule exists and is switched off because it is too noisy.

Purple teaming: attack and defense in the same room

Emulation can run blind, with the SOC unaware of the exercise, to measure real operations. Or it can run as a purple team: the offensive team runs a technique, the defenders watch it happen, and together they tune detection before moving to the next one.

Running blind measures the response as it is today. A purple team speeds up improvement, because many adjustments are tested during the exercise itself; those that need bigger changes come back at the retest. Many companies start with a blind exercise to set a baseline and use purple team rounds to close what stayed open.

An illustrative example

The scenario below is a composite of common findings, not a real customer. A mid-sized insurer chooses to emulate a ransomware operator’s typical behavior.

  • Initial access: an email with a zipped attachment holding a .lnk shortcut (T1566.001). The email gateway delivers it. The EDR records the child process but does not alert.
  • Execution: the shortcut calls PowerShell with a Base64-encoded command (T1059.001). A SIEM rule exists for this, but it filters on a single service account name. It counts as detected, with no alert.
  • Discovery: LDAP queries to Active Directory listing domain accounts (T1087.002) and servers (T1018). No control records them.
  • Credential access: Kerberoasting against a service account with a weak password (T1558.003). The domain controller generates the events, but they are not sent to the SIEM.
  • Lateral movement: the stolen credential used through admin shares (T1021.002). The EDR alerts and the SOC opens a ticket, but hours later.
  • Impact: test files encrypted on a file server (T1486). The EDR blocks it.

The result shows a defense that works at the end of the chain and fails at the start. The fixes come straight from the map: forward the domain controller’s Kerberos events to the SIEM, review the PowerShell rule’s filter, add detection for bulk LDAP enumeration and, where compatible, adopt a managed service account with automatic password rotation. At the retest, the same techniques run again to confirm each adjustment works.

What to ask a supplier

  1. How do you choose the adversary? The answer should cite threat intelligence on your sector and documented groups, not a generic list.
  2. Is the emulation plan delivered before execution? You should see the techniques, their order and the signal expected from each control before anything runs.
  3. How are results classified? Ask for a technique-by-technique map that separates not seen, detected and alerted.
  4. What safeguards protect production? Ask how impact techniques are emulated without destroying data, and who can stop the exercise.
  5. Is there a joint review with the defenders, and a retest? Without a retest, you do not know whether the fixes work.
  6. What does the SOC actually receive? Ideally, recommended rules and use cases, not just a list of what went through.

Next step

Adversary emulation is most useful to companies that already have controls and want to know whether they see what matters. If you do not yet know where an attacker would get in, a pentest or a Red Team can come first. To measure your detection against an adversary that targets companies like yours, see Leonnes’ adversary emulation.

See what an attacker would see first

We reply within one business day.