On this pageRed, Blue and Purple TeamsHow an Exercise RunsThe Working RecordOne Technique From Run to RerunWho Takes PartTelemetry to Confirm BeforehandExercise Safety and CleanupExercise Cadence and TriggersMeasures of ProgressReporting the ResultsPlanning With MITRE ATT&CKContinuous Purple TeamingExercises in Small and Large TeamsStarting a First ExerciseOffense and Detection With ThreatDefenceFrequently Asked Questions An operator runs a technique, the defenders check what their tools recorded, and the gap is fixed and tested again. The name comes from mixing a red team with a blue team, and in most organizations it describes an exercise or a routine more often than a permanent team. This guide explains how a purple team exercise runs, what its working record contains, one technique followed from run to rerun, who takes part, which telemetry to confirm first, how often to run one, how to measure and report progress, how to plan with MITRE ATT&CK, and how automated testing makes the practice continuous. Red, Blue and Purple TeamsThe colors name roles in a security exercise, and purple describes the red and blue roles working as one group. Red team: The red team is the people authorized to act as the adversary, and our guide to red teaming vs penetration testing covers the exercises it runs on its own. Blue team: The blue team is the defenders, described in the NIST glossary as the group responsible for defending an enterprise’s use of information systems by maintaining its security posture against a group of mock attackers. Purple team: The purple team is both groups in one room or on one call, sharing what each can see while the test happens. A penetration test reports which weaknesses in a system can be exploited. A purple team exercise reports, technique by technique, whether the organization’s logging, detections and analysts saw the activity, and it leaves working detections behind. How an Exercise RunsA purple team exercise runs as a loop in which one technique is executed, observed, fixed and repeated before the group moves to the next. Set the goal and scope. The exercise lead agrees with the system owners which hosts and accounts are in play and which question the exercise answers, such as whether the SOC would see an intruder moving from a workstation to a file server. Choose the techniques. The group picks a short list along one plausible attack path, drawn from threat intelligence, a recent incident or an earlier red team report. Write the test plan. Each entry states the procedure, the target, the account, and the log and detection the defenders expect. Prepare the environment. Administrators create test accounts, the sponsor confirms written authorization, and the wider SOC is told how to recognize exercise activity, so a real intrusion is still handled as real. Run one technique. The operator announces the test, executes it, and notes the time, source host, account and command. Check what was recorded. The defenders search for the raw event, then a detection, then an alert that reached an analyst. Record the result. The outcome goes into the working record with the gap named precisely. Fix and run again. A quick fix, such as tuning a rule, is made in the session and the technique is repeated. A larger fix gets an owner and a rerun date. Report. The lead summarizes coverage before and after, the detections added and the items still open. CISA describes the same loop for validating security controls. Its advisory on a red team assessment of a federal civilian agency recommends selecting an ATT&CK technique, testing security technologies against it, analyzing how detection and prevention performed, and tuning people, processes and technologies with the results. Checking for the raw event first separates a logging gap from a detection gap, and the fixes for those usually belong to different people. The Working RecordThe working record is a shared table with one row per test, filled in while the exercise runs. The next exercise starts from it. Technique Expected detection What was seen Gap Fix T1110.003 Password Spraying An alert on failed sign-ins from one source across many accounts. Failures appeared as event 4625 on the server that received the attempts. No alert fired. The rule counted failures per account, and each account failed only once. The rule now counts distinct accounts per source address. The rerun raised an alert. T1003.001 LSASS Memory An endpoint alert on a process opening lsass.exe with full access. The endpoint tool blocked the dump and raised an alert. No case was opened. The alert went to a low-priority queue. Severity was raised and the alert now routes to the on-call analyst. T1053.005 Scheduled Task Event 4698, “A scheduled task was created”, forwarded to the SIEM and matched by a rule. Nothing reached the SIEM. The Audit Other Object Access Events subcategory was off on servers, so the event was never written. The subcategory was enabled by policy. The rerun produced the event and the alert. T1114.003 Email Forwarding Rule An alert on a New-InboxRule operation in the Microsoft 365 audit log that forwards mail outside the organization. The alert fired and an analyst opened a case. None. The test is kept as a regression check. T1567.002 Exfiltration to Cloud Storage A proxy alert on a large upload to an unsanctioned storage service. The upload was visible in proxy logs. No alert fired. No detection existed for upload volume by destination. A detection was written, tuned and rerun. Results are easier to compare when every row is graded on the same scale. Result Meaning Usual next step Prevented A control blocked the action. Confirm the block was logged and visible to the SOC. Alerted A detection fired and reached an analyst. Review triage time and the playbook that followed. Logged only The raw event exists and no detection matched it. Write or tune a detection. No telemetry Nothing was recorded anywhere. Enable auditing or collect the log source, then write the detection. Each kind of gap has its own owner, and the fix is made in a form that outlasts the session. Logging gap: An administrator changes audit policy or log forwarding through central policy, so the setting reaches every host of that type and new hosts inherit it. Detection gap: The detection engineer writes or tunes the rule and stores it together with the procedure that proves it works. Routing gap: The SOC manager changes the severity, queue or on-call routing, so an alert that fires reaches a person. Response gap: An analyst who received the alert without clear next steps gets a playbook entry for that technique. Prevention gap: A weakness the technique relied on, such as a short service account password, goes to the system owner as a hardening item. CISA’s federal agency assessment recorded a routing gap of this kind: the defenders’ daily procedures did not always include analysis of endpoint detection and response alerts. One Technique From Run to RerunFollowing one technique through a session shows how the loop and the record work together. The technique here is Kerberoasting, ATT&CK ID T1558.003, in which an intruder holding a valid domain account requests Kerberos service tickets and attacks them offline to recover service account passwords. The plan entry. The operator will request tickets for the accounts that have a service principal name, from a test workstation, signed in as a standard test account, and will discard the tickets without cracking them. The defenders expect event 4769, “A Kerberos service ticket was requested”, from the domain controllers, and an alert from a rule that looks for ticket encryption type 0x17, the older RC4 cipher. The first run. At 10:02 the operator announces the test and runs it. Within a minute the test account has requested tickets for 14 service accounts. The check. The analysts find all 14 events in the SIEM, with the test account in the Account Name field and the workstation in Client Address. No alert fired, so the row is graded logged only. The gap. Every ticket was issued with encryption type 0x12, which is AES256, because RC4 had since been disabled for those accounts. The rule was written before that change and matched RC4 only. The fix. The detection engineer adds a rule that counts the distinct service names one user account requests within a few minutes. The rerun. At 10:40 the operator repeats the procedure. The alert fires at 10:41 and an analyst opens a case at 10:47, which moves the row to alerted with one minute to alert and six to triage. A second procedure. The operator then requests one ticket every few minutes, which stays under the new threshold, and nothing fires. The administrators create a decoy service account that no application uses, the engineer adds a rule for any ticket request naming it, and the next slow run raises an alert. The follow-up. Both procedures are stored as regression tests. One hardening item stays open for the administrators, since MITRE’s mitigations for this technique include long, complex service account passwords and group managed service accounts. Microsoft’s documentation for event 4769 advises monitoring for a ticket encryption type other than 0x11 and 0x12, the AES values, and for ticket requests naming a service account that the requesting account should never use. Who Takes PartA purple team exercise needs the people who can run a technique, the people who can see it, and the people who can change what is seen. Exercise lead: The lead owns the plan, keeps the pace and maintains the working record. Red team operator: The operator executes each technique and explains how a real adversary would vary it. SOC analysts: The analysts search the telemetry and handle alerts as they would on a normal shift. Detection engineer: The engineer writes and tunes detections during the session. System and identity administrators: The administrators change audit policy, log forwarding and endpoint settings, and provide test accounts. Managed SOC provider: The provider joins as the blue team, or as part of it, where monitoring is outsourced. Sponsor: The sponsor is a security lead who approves the scope and owns the open items afterward. In a small team one person may hold several of these roles, and the operator can come from inside the organization. CISA’s assessment found that the agency’s SOC was unable to update or deploy rulesets quickly through separate IT teams, and a session with the administrators present removes that delay. Telemetry to Confirm BeforehandAn exercise tests detections, so the events those detections read have to exist before the first technique is run. Several of the records that common procedures depend on are controlled by separate settings, and each is worth checking on the hosts in scope. Event What it captures Setting that enables it Event 4688, process creation A new process and the account that started it. The command line appears only when a second setting is on. The Audit Process Creation subcategory, plus the policy “Include command line in process creation events”. Event 4104, PowerShell script block logging The content of the script blocks that PowerShell processes. The Group Policy setting “Turn on PowerShell Script Block Logging”. The events go to the PowerShell operational log. Event 4769, Kerberos service ticket request The requesting account, the service name, the client address and the ticket encryption type. The Audit Kerberos Service Ticket Operations subcategory. The event is generated only on domain controllers. Sysmon event 10, ProcessAccess One process opening another, which is how credential theft tools read the memory of lsass.exe. A Sysmon configuration with filters for expected access, since Microsoft notes the event can otherwise generate significant amounts of logging. Sysmon is the System Monitor service in Microsoft’s Sysinternals tools, and an endpoint detection tool may supply equivalent records. Best practices for event logging and threat detection, published in August 2024 by ASD’s ACSC with CISA, the FBI, the NSA and other partner agencies, recommends always enabling detailed logging that includes process creation events and command-line auditing. For enterprise networks it ranks critical systems, internet-facing services, and identity and domain management servers as the first log sources to collect. A short check of each log source before the session saves time during it. Presence and completeness: A search for each event from each host in scope returns recent records in the SIEM, with the fields a detection reads populated. Forwarding delay: The time between an event and its arrival in the SIEM is known, since it sets a floor under the time to alert. Clocks: Hosts and the SIEM agree on the time, and the joint guidance recommends synchronized time servers with Coordinated Universal Time as the preferred standard. Exercise Safety and CleanupAn exercise on production systems is kept safe by the test plan, which states what each procedure may touch and how it is undone. Written authorization and rules of engagement work as they do for a penetration test. Test accounts and decoy targets: Procedures use accounts, mailboxes and files created for the exercise wherever the technique allows, so real credentials and records stay out of the operator’s hands. Least harmful procedure: The operator picks the mildest action that still produces the telemetry, such as requesting service tickets and discarding them. Controls in audit mode: Where a control blocks a procedure, the group may switch it to audit mode on one host to test the detection behind it, and the change is recorded and reversed. Exercise markers: The SOC receives the source hosts, accounts and time windows, so an analyst on shift can recognize exercise activity and treats everything else as real. Stop conditions: The plan names who can pause the session and what triggers a pause, such as an unexpected outage or signs of a genuine intrusion. Cleanup: Scheduled tasks, inbox rules, test accounts, dropped files and tools are removed at the end and checked off against the operator’s log. Exercise Cadence and TriggersA workable rhythm is a planned exercise at least once a year, with shorter sessions whenever the environment or the detections change. Australia’s CORIE framework, one of the regulator-led programs described in our guide to red teaming vs penetration testing, says institutions in its tiers 1 to 3 should complete a purple exercise annually. The CISA advisory cited earlier goes further and recommends continually testing a security program, at scale and in a production environment, against the ATT&CK techniques it identifies. Certain events justify a session of their own. A platform change: After a SIEM migration or a new identity provider or endpoint tool, detections that worked before are rerun. New or retuned detections: A rule is tested against the real technique before it is trusted. An incident: The techniques the intruder used are replayed once the fixes are in place. A red team or penetration test report: The attack path in the report becomes the technique list. A relevant advisory: Techniques reported against the same sector or the same products are tested soon after publication. Measures of ProgressProgress shows in the working record over time, as more of the tested techniques are seen, sooner, and stay seen. Coverage by result: The record gives the share of tested procedures that were prevented, alerted on, logged only or left without telemetry. Time to alert: The interval runs from the operator’s execution time to the alert. Time to triage: The interval runs from the alert to an analyst’s decision. Gaps closed: The count covers the open rows fixed and rerun, and how long each stayed open. Regressions: A regression is a test that passed in an earlier exercise and fails now. Behavior-based detections: The measure is the share of detections keyed to what the adversary does, as opposed to a file name, hash or address that changes with the next tool. CISA’s federal agency assessment shows why the last measure matters. In its collaboration phase the red team held weekly discussions with the agency’s SOC and administrators and withheld specific indicators, so the defenders built detections around behavior. CISA reported measurable improvements in response times for known techniques. Full coverage of a short, easy list says little, so each figure is reported together with the techniques and procedures it covers. Reporting the ResultsThe report turns the working record into a short document for the people who own and fund the fixes. Scope and method: The report lists the hosts, accounts and dates, the techniques and procedures run, and who took part. Coverage before and after: Each procedure appears with its grade at the first run and at the last rerun. Detections added or changed: Each rule is named with the technique it covers and the procedure that proved it. Open items and reruns: Every unresolved gap has an owner and a due date, and the stored tests have a rerun schedule. Each reader takes something different from it. The security lead reads the change in coverage and chooses the next technique list, the IT manager schedules the audit policy and forwarding changes, and the SOC manager follows the routing and playbook items. The report can also serve as evidence that detection is tested. The NIST Cybersecurity Framework 2.0 includes the outcome ID.IM-02, under which improvements come out of security tests and exercises, and its Detect function covers continuous monitoring and the analysis of adverse events. In an exposure management program the same results belong to the validation stage, described in our guide to continuous threat exposure management. Planning With MITRE ATT&CKMITRE ATT&CK gives the plan a shared vocabulary, so the operator’s action, the detection rule and the report all refer to the same technique ID. MITRE describes ATT&CK as a globally accessible knowledge base of adversary tactics and techniques based on real-world observations. Its common uses, as MITRE lists them, include adversary emulation and red teaming, detections and analytics, and assessment and engineering. Tactic: A tactic is the reason for an action, such as Credential Access or Lateral Movement. Technique and sub-technique: A technique is how the adversary achieves that goal, and a sub-technique describes the behavior more specifically. Procedure: A procedure is the specific implementation an adversary uses, which is what the operator runs. Several inputs feed the technique list. Input What it gives the plan Where to find it ATT&CK technique pages A description, procedure examples, mitigations and a detection strategy, which groups platform-specific analytics for the technique. The Techniques section of the ATT&CK website. ATT&CK group pages The techniques publicly reported for a named adversary group, with references. The Groups section of the ATT&CK website. Government advisories Techniques observed in a real intrusion or assessment, listed in tables of technique IDs. CISA cybersecurity advisories, including its red team assessment reports. Detection inventory The techniques the existing rules claim to cover. The SIEM. Microsoft Sentinel, for example, has a MITRE ATT&CK page showing the analytics rules active in a workspace by technique. ATT&CK Navigator A layer that marks each technique as planned, tested or detected. MITRE’s Navigator tool, built to visualize defensive coverage and support red and blue team planning. A coverage map built from rule mappings shows where detections exist, and the exercise shows whether they fire. One technique can be carried out through many procedures, so a detection that catches one tool may miss another. The plan should hold a second procedure for the techniques that matter most. MITRE updates ATT&CK twice a year, so technique IDs in an older record are worth checking against the current version before a rerun. Continuous Purple TeamingContinuous purple teaming keeps the loop running between exercises by automating the attack side and rerunning stored tests on a schedule. Each row in the working record becomes a test case with a known expected result, and a failed rerun shows that a detection has drifted. Scripted technique tests: Small scripts reproduce one procedure each on agreed hosts, and MITRE Caldera, an open-source platform for automated adversary emulation built on ATT&CK, is one tool used for this. Breach and attack simulation: Products run libraries of simulated techniques against deployed controls, a category covered in our comparison of penetration testing vs vulnerability scanning. Agentic testing: AI agents test exposed systems when something changes and keep proof of each confirmed weakness, which gives defenders a timestamped event to look for. Automated reruns confirm what is already known, while people choose new techniques, interpret a partial result and design the fix. Automated tests on production systems keep the same limits as manual ones: test accounts, agreed hosts, no destructive actions and a record of every run. Exercises in Small and Large TeamsThe method stays the same at any size. What changes is who plays each role and how much is tested at once. Small IT team with a managed SOC: An administrator runs a few well-understood procedures from a test workstation while the provider’s analysts watch their platform, and the first sessions mostly find logging gaps. Mid-size security team: An internal engineer or an external tester acts as operator for a few days, a detection engineer attends throughout, and short sessions follow each platform change. Large organization with a red team: Exercises run on a standing schedule from a backlog of techniques, ordered by the adversary groups that matter to the sector, and the regression set runs automatically in between. MSP or MSSP: The provider runs one technique list in each client environment under that client’s authorization, because audit policy and log sources differ between clients and a detection that fires for one may stay silent for another. Where monitoring is split between parties, an exercise also shows who is watching which log. CISA’s assessment found that the agency’s defenders did not monitor command and control traffic leaving over DNS, in the belief that a parent entity was monitoring it. Starting a First ExerciseA first exercise works best when it is small and ends with fixes in place. Confirm the logs first. The team checks that sign-in, endpoint and network logs from the hosts in scope reach the SIEM, since otherwise every row ends with the same gap. Borrow the technique list. The attack path in the last penetration test report, or the technique tables in a government advisory for the sector, gives a first list without new research. Bring the people who can change things. With a detection engineer and an administrator in the session, a quick fix is made and rerun the same day. Schedule the reruns before closing. The next session begins by rerunning the items left open. Offense and Detection With ThreatDefenceThreatDefence links offensive testing to detection for internet-facing systems. Both sides run in one AI SOC platform, backed by a 24x7 SOC, that automates monitoring, compliance and the management of cyber risk. A weakness is confirmed. In agentic penetration testing, the testing agent shows that a weakness on an exposed system is exploitable, uses the least intrusive proof, and logs every request and result. The defensive agents take it up. For a monitored customer, the confirmed weakness goes to the Defender and Hunter agents, which search the customer’s telemetry for signs that someone used it earlier. Analysts and retests close it. SOC analysts review significant findings, and the testing agent tests again once a fix is reported. This is automated testing with a detection check, and it stands apart from a purple team exercise. An exercise on internal techniques, such as credential dumping and lateral movement, stays with the organization’s own team or an independent human-led provider, and it can run alongside. Frequently Asked QuestionsIs a purple team a separate team?No, in most organizations. Purple teaming usually describes red team operators and defenders working together for the length of an exercise, then returning to their normal roles. Some larger organizations keep a small standing group for it, and the work is the same either way: run a technique, check what was seen, fix the gap and run it again.How long does a purple team exercise take?It depends on the technique list. A focused session on a few techniques along one attack path can fit into a single working session. A formal exercise that replays a full red team attack path, fixes the gaps and ends with a written report takes considerably longer and is planned as a project, and short sessions can run between larger exercises.Does purple teaming replace a red team exercise?No. A red team exercise tests whether the defenders notice an unannounced attack under realistic conditions, and purple teaming removes that surprise on purpose so detection can be built quickly. The usual order is purple teaming first, to confirm that logging and detections work, and a red team exercise later to test the whole defense.Can an internal team run a purple team exercise?Yes. The method needs someone able to execute techniques safely, access to the SIEM and endpoint tools, and administrators who can change logging, and all of those can come from inside the organization. An independent red team provider is optional. An external operator may still add procedures the internal team has never tried, which is the usual reason to bring one in.Does a purple team exercise run on production systems?Yes, in most cases, because the aim is to test the logging and detections that are actually deployed. CISA recommends testing a security program at scale in a production environment. The exercise uses test accounts and agreed hosts, avoids destructive actions, and runs under written authorization, with the SOC told how to recognize exercise activity.Is MITRE ATT&CK required for purple teaming?No. An exercise can be planned from an incident report or a red team’s operator log without any framework. ATT&CK helps because it gives each technique a stable ID and a description, so the operator’s action, the detection rule and the working record refer to the same thing, and results from different exercises can be compared.What does a purple team exercise produce?It produces a working record with one row per test, showing the technique, the expected detection, what was seen, the gap and the fix. It also leaves new or tuned detections, changes to logging, a list of open items with owners, and a set of tests that can be rerun later to confirm that nothing has regressed.Can purple teaming work with an outsourced SOC?Yes. The managed SOC provider takes part as the blue team, or as part of it, searches its own platform for each technique and tunes detections during the session. The organization’s administrators still attend, because many fixes are changes to audit policy or log forwarding on systems the provider does not manage.How is purple teaming different from a tabletop exercise?A tabletop exercise is discussion-based. NIST describes it as personnel meeting to validate the content of a plan by discussing their roles and their responses to a scenario that a facilitator presents. A purple team exercise executes real techniques on real systems and checks what the tools recorded, so it tests telemetry and detections where a tabletop tests plans and decisions.What happens if a real attack occurs during a purple team exercise?The session pauses and the activity is handled as a real incident. The SOC is given the exercise’s source hosts, accounts and time windows beforehand, so anything outside them is treated as genuine. The test plan names who can pause the session, and the operator’s log of times and commands helps separate exercise activity from an intruder’s.