The incident response process is the ordered sequence a security team follows from the first suspicious signal to the last documented improvement. This article walks it in seven stages: preparation, detection, analysis and triage, containment, eradication, recovery, and post-incident improvement. The failures concentrate at the seams between them.
What Is the Incident Response Process?
The incident response process is the defined sequence an organization follows to detect, contain, eradicate, and recover from a cybersecurity incident, then to improve from it. It states what happens, in what order, and what each step produces. The incident response plan is the document that records that sequence: a separate artifact with its own contents, approvals, and review cycle. A team can hold a current plan and still run an undefined process. CISO Cyber Incident Response Plan: Prepare, Test & Respond covers the plan as a separate artifact.
Two documents dominate practice. NIST SP 800-61r3, published April 2025, supersedes r2 and maps incident response onto the six Functions of Cybersecurity Framework 2.0 rather than a standalone four-phase lifecycle. What practitioners call the NIST incident response process now sits distributed across those six Functions rather than in a numbered list of its own. The SANS Incident Handler's Handbook (2012) groups the same work into six named steps, which practitioners commonly shorten to PICERL. The handbook itself does not use that acronym. Many operating runbooks are still written against those step names.
What changed in that revision is covered by the revised incident handling guidance analysis on this site. This article focuses on the sequence. The two models cover similar work but draw its boundaries differently.
The 7 Stages of an Effective Incident Response Process
Neither anchor document publishes a seven-stage list. The incident response process NIST describes in SP 800-61r3 spreads across six CSF 2.0 Functions. SANS names six steps, bundling detection, determination, and scoping into a single Identification phase. The seventh stage here comes from splitting that phase in two. Teams staff, queue, and measure the halves apart, and r3 draws the same line: an incident is declared under Detect (DE.AE-08) and triaged under Respond (RS.MA-02). The numbering is this article's. The phases belong to the standards.
r3 also declines to make the choice for anyone. Section 2.1 states it directly: "Organizations should use the incident response life cycle framework or model that suits them best."
Stage | NIST SP 800-61r3 | SANS Incident Handler's Handbook (2012) | What neither document specifies |
1. Preparation | Govern, Identify, Protect. These Functions support incident response rather than forming a response phase | Preparation | How often the artifacts must be re-verified |
2. Detection | Detect. DE.AE-08: incidents declared when events meet defined criteria | Identification | The incident criteria themselves |
3. Analysis and Triage | Respond. RS.MA-02 (triage and validation), RS.MA-03 (categorize and prioritize) | Identification | The severity scale, or how much evidence is enough |
4. Containment | Respond. RS.MI-01 | Containment | When a service outage is an acceptable price |
5. Eradication | Respond. RS.MI-02 | Eradication | How completeness is proven |
6. Recovery | Recover. RC.RP-01 through RC.RP-06 | Recovery | Who declares the end of recovery |
7. Post-Incident Improvement | Identify. Improvement Category (ID.IM), fed continuously from all Functions | Lessons Learned, terminal, aimed at the next incident | Who owns a finding once the report is filed |
1. Preparation
Preparation is the continuous work done before an incident: defining incident criteria, assigning roles, provisioning logging, and rehearsing the sequence. It produces a named on-call rotation, a severity scale with thresholds, an evidence retention window, and a contact tree someone has dialed. Detection inherits the criteria. Without them, an analyst has an alert and no basis to call it anything.
r3 places preparation outside the response itself: Govern, Identify, and Protect support incident handling rather than perform it. The break here is quiet. Every artifact exists, and three are stale. A regional carrier's escalation list still routes to an engineer who moved teams in March.
2. Detection
Detection is the stage where a signal becomes a declared incident. NIST supplies the sentence and leaves the substance to the organization. DE.AE-08 reads: "Incidents are declared when adverse events meet the defined incident criteria." The criteria are the organization's own. Detection produces a timestamped declaration, the initial indicators, and their source.
Triage inherits the declaration and its timestamp. Mean time to detect ends at that mark, while later response intervals begin from it. The stage breaks when the signal arrives and nobody declares. At the carrier, an anomalous authentication pattern against an OSS/BSS administrative interface reads as a broken integration for two shifts. It has always been a broken integration before.
3. Analysis and Triage
The analysis and triage stage validates a declared incident, scopes it, and sets a severity. r3 splits the work across two Subcategories: RS.MA-02 covers triage and validation, including an estimate of severity and urgency; RS.MA-03 covers categorization and prioritization. It produces a severity, a scope statement, a working hypothesis, and a first assessment of whether a notification clock has started.
Containment inherits the severity, because severity is commonly what grants containment authority. This stage breaks on evidence. Severity gets set from what is visible in the first twenty minutes, and rarely revisited. At the carrier, the anomaly is scoped as an access-control problem. Subscriber data enters the scope four hours later.
4. Containment
Containment prevents the incident from expanding. r3 states it plainly under RS.MI-01: containment "refers to preventing the expansion of an incident." The stage produces a boundary: an isolated segment, a revoked credential set, or a suspended account, plus a record of what was cut, when, and by whom.
Eradication inherits that boundary and whatever evidence survived it. This stage carries a cost the first three do not. A carrier cannot isolate a signaling segment without dropping subscriber traffic, and the engineer at the console cannot authorize that outage. The break is structural. The decision waits for authority, and the incident moves while it waits.
5. Eradication
Eradication removes the attacker's access and the conditions that permitted it. r3 describes it under RS.MI-02 as mitigating an incident's effects after containment: deleting malware, disabling breached accounts, remediating exploited vulnerabilities. It produces a list of remediated hosts, closed accounts, and patched flaws, plus an implicit claim that the list is complete.
Recovery inherits the completeness claim. Neither model defines how completeness is proven. A team eradicates what it found, and the scope set in stage 3 capped what it found. When triage undercounted, eradication looks finished. It is not.
6. Recovery
Recovery restores affected services and verifies restored systems are clean before they carry production traffic. r3 gives this stage its most granular treatment: RC.RP-01 through RC.RP-06 cover plan execution, selection of recovery actions, backup integrity, post-incident operational norms, verification of restored assets, and the declared end. RC.RP-06 reads: "The end of incident recovery is declared based on criteria, and incident-related documentation is completed." It names neither the criteria nor the person who applies them.
Improvement inherits the documentation. Pressure to restore service peaks here. A team under it declares it done before eradication has been proven, and reinfection follows. The eradication-recovery boundary is the cause.
7. Post-Incident Improvement
Post-incident improvement turns what happened into a change somebody owns. The two models diverge most here. SANS calls it Lessons Learned and places it last, with a stated goal of improving the team's performance and producing reference and training material for the next incident. r3 relocates the work: improvements live in the Improvement Category (ID.IM) inside Identify, fed continuously from all six Functions rather than only after recovery ends. ID.IM-02 names exercises as one source: "Improvements are identified from security tests and exercises, including those done in coordination with suppliers and relevant third parties."
The stage produces an after-action report, findings, and, at best, an owner beside each one. It breaks on the owner column. A finding with no name attached is a sentence in a document.
See what your team does under real pressure.
Put your readiness to the test. Sixty minutes, your real environment, no slides.
Book A demo
Who Does What During a Cyber Incident?
A cyber incident response process assigns a primary actor to each stage and a secondary actor who supplies what the primary cannot. r3 declines to publish a fixed cast: roles differ by organization, and within one organization by incident. The map below names who acts, in sequence. Decision rights are covered in Incident Response Roles and Responsibilities: Who Owns What?.
Preparation. Primary: the incident response lead. Secondary: IT operations and the control owners who maintain logging.
Detection. Primary: the SOC analyst on shift. Secondary: the detection engineer who owns the rule.
Analysis and triage. Primary: the incident commander. Secondary: the threat analyst and the system owner.
Containment. Primary: the incident commander. Secondary: network and platform engineering, who execute.
Eradication. Primary: security engineering. Secondary: the system owner.
Recovery. Primary: IT operations. Secondary: security engineering, verifying integrity.
Post-incident improvement. Primary: the incident response lead. Secondary: the named owner of each finding.
Legal, Communications, and executive leadership enter at triage, when severity is set, and stay to the end. Their coordination under time pressure is covered in Incident Response Management: From Detection to Crisis Recovery.
The cast changes six times in seven stages. Each change is a handoff.
Where Incident Response Processes Break Down Under Pressure
Incident response processes break at the boundaries between stages, not inside them. Each stage has an owner, a tool, and a measurement. The transfer between two stages has none of the three. Five break points recur, and all five sit on a seam.
The detection-to-triage handoff. An analyst escalates into a queue measured on volume. Mean time to detect stops at the declaration. Nothing starts at the other end, so the interval before the first triage decision belongs to no metric and no person. Incident Response Metrics: 10 KPIs That Actually Measure Readiness explains how to measure that gap.
The severity call with incomplete evidence. Severity gates containment authority, and severity gets set before the evidence is in. Set it low and the response stays under-resourced for hours. Set it high and the team burns credibility it needs next time. Neither standard supplies the threshold.
Containment that costs service availability. The carrier case again: the isolation is technically trivial and organizationally expensive. The engineer with the console cannot authorize dropping subscriber traffic. The executive who can is not on the bridge. The incident expands during a phone call.
The eradication-recovery boundary. A team declares eradication complete because the indicators went quiet, and recovery treats that declaration as fact. The verification step that would catch it, RC.RP-05, sits after the boundary rather than before it.
Improvement findings nobody owns. Exercises and real incidents both produce findings into documents with no owner column. This is the defect described in no failure condition: an activity that cannot be failed generates paper rather than change.
This is the pattern Reflex Security watches most closely in live simulations. The stage a team has rehearsed is rarely the stage that fails. Some incidents outgrow the sequence entirely, at which point the question stops being what happens next and becomes who decides.
Documentation prevents none of the five. Each is a transfer. A transfer exists only when someone is waiting on the other side.
How to Test Your Incident Response Process Before a Breach
Testing an incident response process means putting the sequence under a clock and finding where it stalls. Three methods are in common use. Each proves something real, and each leaves specific stages untested. Common control frameworks ask whether you tested the security incident response process. They do not ask which of the seven stages you tested.
The runbook walkthrough proves the document is current, reachable, and internally consistent. It catches the stale contact tree from stage 1, which is worth the hour on its own. It cannot test a handoff. Nobody is waiting at the other end of one.
The tabletop exercise proves participants can describe the sequence, and it surfaces disagreement about who acts when. That is the real value: it is the fastest way to find two people who each believe the other declares. Without a clock or a consequence, it cannot prove an escalation happens under load. The argument for moving beyond static exercises rests on that limit.
The purple-team exercise proves detection and containment against real technique, at real speed, with real telemetry. It stops at the technical boundary. Severity assignment, the notification assessment, the authority to accept an outage, and the recovery declaration go untested. Four stages of seven.
A process test measures time to each handoff rather than time to resolution. Mean time to resolve says the run was long. It does not say which seam absorbed the time. The findings then have to survive the week, which is the harder problem and the subject of turning findings into change.
A stage that has never been timed has never been tested.
How Reflex Security Turns Your IR Process Into a Live Simulation
The process needs four things a document cannot supply: an escalation threshold that has been exercised, a handoff that has been timed, a containment decision made on incomplete information, and a finding that outlives the meeting. Reflex builds an exercise around each one.
An escalation threshold that has been exercised. A severity scale is a hypothesis until someone applies it to a scenario they have not read in advance. Reflex One-Click Scenario Generation returns five customized scenarios in 10 to 15 minutes, built from OSINT about your organization: its domain, its documented technology, its named executives. The threshold gets tested against the environment it was written for.
A handoff that has been timed. The interval between a declaration and a triage decision is invisible in a discussion and decisive in an incident. Reflex logs every decision, escalation, and query as it happens, with the Investigation Console as the surface where participants work the evidence in plain English. An AI facilitator named Victoria runs the session and challenges the calls across an exercise of 60 to 120 minutes. The gap becomes a number instead of an impression.
A containment decision made on incomplete information. A scripted inject tells the team what happened, which is why scripted injects cannot test stage 4. The Adaptive Adversary Engine reacts to what the team does instead, so a containment call made on partial evidence produces a consequence the team then has to handle. A human facilitator can steer the same run through the Reflex Facilitator Guide. The mechanics of adaptive scenario generation are described separately.
A finding that outlives the meeting. The Reflex After Action Report (AAR) typically exceeds 54 pages and carries the timestamps with it, so each finding arrives attached to the minute it happened and the decision that caused it. That is what an owner needs.
The seven stages are the process. What happens to them at minute forty of a run nobody scripted is the evidence.
Start with one scenario built on your own domain. Book a demo.
Frequently Asked Questions
Should runbooks be rewritten against NIST SP 800-61r3?
Not urgently. r3 states that organizations should use the life cycle model that suits them best, so a runbook written against the SANS step names stays valid. What is worth doing is mapping your existing steps to the CSF 2.0 Subcategories, because auditors and CSF-aligned reporting increasingly ask in that vocabulary.
How many severity levels should a severity scale have?
Three or four that a tired analyst can apply at 3 a.m. Neither NIST nor SANS sets a number. What breaks a scale is not its size but its thresholds: if two people reading the same evidence land on different levels, the scale is describing a feeling rather than a condition.
What are reasonable target times for each stage?
Targets belong to your environment, and borrowed benchmarks mislead. Time the handoffs for a quarter first, then set targets from your own distribution. The interval worth instrumenting before any other is declaration to first triage decision, because most organizations have never measured it at all.
Does an MDR provider run the incident response process for you?
It runs part of it: detection, much of triage, and often the first containment actions. The stages it cannot hold are the ones requiring your authority. Accepting a service outage, setting severity against business impact, declaring recovery complete. Read the service description against the seven stages and mark which ones come back to you.
Does a small team need all seven stages?
Every stage happens whether or not it is named, and the question is who holds it. On a three-person team one person carries detection, triage and containment, so the handoffs become internal rather than disappearing. Name them anyway. An unnamed handoff inside one person's head is still where the time goes.
How often should the incident response process be reviewed?
After every significant incident, after every exercise that produced a finding, and after any change to the on-call rotation, the severity scale or the escalation chain. A calendar review once a year catches stale contact details and little else, because the artifacts that age fastest are the ones people change quietly.
References
National Institute of Standards and Technology. (2025, April). Incident response recommendations and considerations for cybersecurity risk management: A CSF 2.0 community profile (NIST SP 800-61r3). NIST Computer Security Resource Center. https://csrc.nist.gov/pubs/sp/800/61/r3/final
Kral, P. (2012). Incident handler's handbook. SANS Institute. (GIAC GCIH Gold paper, accepted 5 December 2011.) https://www.sans.org/white-papers/33901/
National Institute of Standards and Technology. (2024). The NIST cybersecurity framework (CSF) 2.0. National Institute of Standards and Technology. https://www.nist.gov/cyberframework
Goldschmidt, C. (n.d.). Cassio Goldschmidt: Contributor profile. CSO Online. https://www.csoonline.com/profile/cassio-goldschmidt/
Goldschmidt, C. (n.d.). Cassio Goldschmidt: Contributor profile. MSSP Alert. https://www.msspalert.com/contributor/cassio-goldschmidt
Reflex Security. (n.d.). AI-powered tabletop exercises for incident response. https://reflexsecurity.io/blog/how-ai-powered-tabletop-exercises-improve-incident-response-readiness
Subcategory identifiers (DE.AE-08, RS.MA-02, RS.MA-03, RS.MI-01, RS.MI-02, RC.RP-01 through RC.RP-06, ID.IM-02) and all quoted text were verified against the published SP 800-61r3 PDF. The six-phase sequence and the wording of the Lessons Learned phase were verified against the published SANS paper. Product figures and component names were verified against Reflex Security's published platform description. Verification date 2026-09-10.
