tonia.ai · Concepts

Incident Response and Management

Every prevention control eventually fails. Incident response is the discipline of noticing quickly, limiting the damage, removing the attacker, restoring service, and learning enough that the same failure does not happen twice. It is decided long before the incident starts.

01Events, alerts, and incidents

A network produces millions of events a day: logins, connections, file reads. A small fraction trigger alerts because they match a rule. Only some alerts, after investigation, turn out to be incidents: actual or imminent violations of security policy that threaten confidentiality, integrity, or availability.

Eventsmillions per day · logs, flows, telemetry Alertshundreds · matched a detection rule Incidentsa few · confirmed harm triage declare
Formally declaring an incident matters: it starts the clock, activates the response team, and triggers legal and reporting obligations.

high Severity

How bad is the impact? Data exposure, outage, safety, regulated information.

scope Breadth

One laptop, one department, or the domain controllers and backups?

time Urgency

Is the attacker still active? Is data leaving now? Is a deadline running?

02The incident response lifecycle

The classic model has four phases arranged as a loop, because every incident improves preparation for the next. Recent guidance places this loop inside broader risk management: governing, identifying, and protecting come before any incident, and response feeds back into them.

Preparationplans · tools · people Detection & analysistriage · scope · declare Contain · eradicate · recoverstop spread · remove · restore Post-incidentlessons · improvements iterate as new hosts are found
Detection and containment cycle repeatedly: each containment step reveals more of the intrusion, which reopens analysis.

03Preparation: deciding before the crisis

The quality of a response is mostly set before the incident: what was logged, who has authority, which contacts are known, and whether anyone has practiced.

Plan & authority

A written plan defining the incident team, who can declare an incident, who can take systems offline, and when to involve legal, executives, and law enforcement.

Playbooks

Step-by-step runbooks for common cases: phishing, ransomware, lost laptop, compromised account, DDoS.

Visibility

Centralized, time-synchronized logs kept long enough to look back months. Endpoint telemetry. Network flow records. You cannot investigate what you never recorded.

Out-of-band channels

A way to communicate if email, chat, and identity systems are compromised or encrypted.

Asset knowledge

An inventory of systems, owners, and data sensitivity, so responders know what matters and whom to call.

Exercises

Tabletop drills that walk leaders through a scenario. They reveal missing contacts, unclear authority, and untested backups cheaply.

04Detection and analysis

Most incidents are discovered through a mix of security tooling (SIEM, EDR, IDS), user reports, and, too often, outside parties such as a customer, a researcher, or the attacker's ransom note. Analysis turns an alert into an answer to four questions: what happened, which systems, since when, and is it still happening?

email gatewayphish delivered macro runsEDR firewall / DNSbeacon to C2 admin loginauth logs netflowlarge upload detected dwell time = first malicious event → detection; every source must share an accurate clock
The timeline is the central artifact of an investigation. Gaps in it usually mark missing logs, not missing attacker activity.

Analysts fight two problems at once: alert fatigue, where real signals drown in false positives, and scoping, where the first compromised host found is rarely the first one compromised. Correlating evidence across hosts, accounts, and network flows is what reveals the full intrusion.

05Models for understanding an intrusion

Responders use attack models to organize evidence, predict the attacker's next step, and choose where to act.

Recon Weaponize Deliver Exploit Install Command& control Actions onobjectives break the chain early → cheaper later → more damage done Intrusion kill chain
Stopping any one stage defeats the campaign; analyzing all stages reveals where defenses failed.

Kill chain

A linear sequence of attacker stages. Useful for asking "where could we have stopped this?" Criticized for focusing on the perimeter.

MITRE ATT&CK

A detailed matrix of observed tactics and techniques. Maps evidence to behaviors and highlights detection gaps.

Diamond model

Links adversary, capability, infrastructure, and victim for each event, so one clue (an IP, a malware family) pivots to others.

Pyramid of Pain

Ranks indicators by attacker cost to change: hashes and IPs are trivial, tools and behaviors (TTPs) are expensive. Detect high on the pyramid.

06Containment

Containment stops the bleeding while preserving the ability to understand what happened. It is a trade-off between speed, evidence, business impact, and tipping off the attacker.

less disruptive · attacker may not notice more disruptive · attacker will notice Monitorwatch & scopebefore acting Block indicatorsC2 domains, IPs,hashes Isolate hostsEDR networkquarantine Cut segmentsdisconnect sites,reset all credentials
Acting before the scope is known lets the attacker retreat to footholds you have not found yet. Waiting too long lets damage grow. Judgment, not a formula, sets the timing.

Avoid pulling the power on a compromised machine unless damage is imminent: memory holds running processes, network connections, and encryption keys that disappear at shutdown. Network isolation preserves that evidence while stopping communication.

07Evidence and forensics

If an incident may lead to legal action, insurance claims, or regulatory scrutiny, how evidence is collected matters as much as what it shows. Two principles guide collection:

Order of volatility: collect top first CPU registers, cache memory, network connections running processes, temp files disk remote logs, backups seconds → years Chain of custody collect · hash (SHA-256) · label log every handoff: who, when, why analyze copies, never the original matching hashes prove nothing changed
Collect the most perishable evidence first, and be able to prove it was not altered afterward.

08Eradication and recovery

Eradication removes every attacker foothold: malware, backdoor accounts, persistence mechanisms, stolen credentials, and the original vulnerability. Recovery returns systems to service in a trustworthy state and watches them closely for re-entry.

Rebuild, don't clean

A compromised host can hide persistence anywhere. Reimaging from known-good media is more reliable than removing what you found.

Reset credentials broadly

Assume any credential on a compromised system is stolen, including service accounts and, in domain compromises, the Kerberos key that signs all tickets.

Fix the entry point

Patch, close the exposure, or add MFA. Otherwise the attacker returns the same way.

Restore and verify

Restore from backups taken before the compromise, then confirm integrity and heightened monitoring before reconnecting.

Removing the attacker in stages tips them off and lets them dig in elsewhere. Plan eradication so that all known footholds are removed at the same time.

09Lessons learned

The post-incident review is where the organization gets value back from the incident. It should be blameless: the goal is to understand why the system allowed the failure, not to find someone to punish, because people who fear blame hide information.

QuestionProduces
What happened, and when? (timeline)Shared factual record
Why did prevention fail? Why wasn't it detected sooner?Control and detection gaps
What slowed the response?Process, tooling, and authority fixes
What would we do differently?Owned, dated action items

Useful metrics include time to detect, time to contain, and time to recover, tracked across incidents. Trends matter more than any single number.

10People, roles, and communication

Incidents are as much organizational events as technical ones. A clear structure prevents the most common failure: many people working hard in uncoordinated directions.

Incident leaddecides · coordinates Technicalanalysts · IT ops Legal & privacyobligations · privilege Communicationsstaff · customers · press Leadershipbusiness decisions plus external: insurer, outside forensics firm, law enforcement, regulators
One person coordinates; specialists work their tracks; a scribe keeps a timestamped log of decisions.

Many jurisdictions and sectors impose deadlines for notifying regulators or affected individuals, some measured in days or even hours. Responders need to know which apply to their organization before an incident, not discover them during one.

11Ransomware: a worked example

Ransomware compresses every phase into hours. Modern operators steal data before encrypting and threaten to publish it ("double extortion"), so restoring from backup no longer ends the incident by itself.

PhaseKey actions
PreparationOffline or immutable backups, tested restores, MFA on remote access, a ransomware playbook
DetectionWatch for precursors: credential dumping, disabled security tools, mass file renames, backup deletion
ContainmentIsolate affected segments, disable compromised accounts, protect backups first
EradicationFind the initial access and all persistence; rebuild domain trust if domain controllers were touched
RecoveryRestore by business priority from clean backups; monitor for re-encryption
Post-incidentAssess data exposure and notification duties; close the entry point everywhere

Whether to pay is a business, legal, and ethical decision made with counsel. Payment does not guarantee a working decryptor or deletion of stolen data, and paying certain sanctioned groups is illegal in some jurisdictions.

12Common misconceptions

MythIncident response is an IT task.
RealityLegal, communications, leadership, and HR decisions often determine the outcome as much as the technical work.
MythShut everything down immediately.
RealityPremature action destroys evidence and alerts the attacker before you know the full scope.
MythOnce the malware is removed, we're done.
RealityStolen credentials, backdoor accounts, and the original vulnerability remain unless deliberately eradicated.
MythWe'll write the plan when we need it.
RealityDuring an incident there is no time to discover who has authority or where the logs are.

—References

Peer-reviewed

  1. Ahmad, A., Hadgkiss, J. & Ruighaver, A. B. (2012). Incident response teams: challenges in supporting the organisational security function. Computers & Security, 31(5). doi:10.1016/j.cose.2012.04.001
  2. Alahmadi, B. A., Axon, L. & Martinovic, I. (2022). 99% false positives. USENIX Security. usenix.org
  3. Garfinkel, S. (2010). Digital forensics research: the next 10 years. Digital Investigation, 7. doi:10.1016/j.diin.2010.05.009
  4. Hassan, W. U. et al. (2019). NoDoze. NDSS. doi:10.14722/ndss.2019.23349
  5. Hutchins, E., Cloppert, M. & Amin, R. (2011). Intelligence-driven computer network defense. ICIW. lockheedmartin.com
  6. King, S. T. & Chen, P. M. (2003). Backtracking intrusions. ACM SOSP. doi:10.1145/945445.945467
  7. Kokulu, F. B. et al. (2019). Matched and mismatched SOCs. ACM CCS. doi:10.1145/3319535.3354239
  8. Sundaramurthy, S. C. et al. (2014). An anthropological approach to studying CSIRTs. IEEE Security & Privacy, 12(5). doi:10.1109/MSP.2014.84
  9. Tøndel, I. A., Line, M. B. & Jaatun, M. G. (2014). Information security incident management. Computers & Security, 45. doi:10.1016/j.cose.2014.05.003

Standards and guidance