Field notes, not a textbook. The NIST incident handling guide (SP 800-61) draws the lifecycle as a loop for a reason. Revision 2, the long-standing version, has four phases: preparation, detection and analysis, containment, eradication and recovery, and post-incident activity. A later revision (Rev 3, 2025) re-maps the same work onto the functions of the NIST Cybersecurity Framework 2.0. Other shops use the SANS six-step version, which just splits the middle phase into three. The labels differ. The work does not.

Preparation decides the outcome

Most of an incident is won or lost before it starts. Opinion, but a firm one: if on the day you are still asking who has authority to pull a production server off the network, you have already lost hours.

What preparation means in practice:

  • A written plan with named roles. An incident commander, someone for communications, someone for legal, someone who can talk to the cloud provider.
  • A contact list that works at 3 a.m. and does not live only on the systems that are down.
  • Logs that exist and are kept long enough. You cannot investigate what was never recorded.
  • Pre-agreed decisions: who may isolate a business-critical system, and at what severity.
  • Practice. A table-top exercise twice a year finds more gaps than any audit.

Detection and analysis: be sceptical in both directions

An alert is a hypothesis. Your job is to confirm or kill it quickly. Two failure modes show up again and again. One is dismissing real signals because the last forty alerts were false. The other is declaring a major incident on thin evidence and burning the team’s credibility.

Write down what you know, what you suspect, and what you do not know, separately. Keep a running timeline with timestamps in one time zone. Use IST or UTC, but pick one and say which. Mixed time zones across cloud logs, endpoint tools and email headers have wrecked more timelines than any clever attacker.

Classify by impact, not by how exciting the malware is. A boring credential leak on a payroll system can matter more than a noisy worm in a test lab. Define severity levels in advance, tied to data and business function.

Containment: the tension nobody mentions early enough

Containment stops the spread. Evidence preservation records what happened. They conflict. Pull the plug and you lose what was in memory. Leave the machine running to watch the attacker, and the attacker keeps encrypting files.

My rule: for fast-moving damage, such as ransomware, contain first and accept the evidence loss, but capture what you can in the minute it takes (memory image, disk snapshot in cloud environments). For slow, quiet intrusions, take time to scope before acting, because a half-hearted cleanup tips off the attacker and they return through a door you never found.

Containment comes in short-term and long-term flavours. Short-term: isolate the host, disable the account, block the address. Long-term: rebuild the segment, rotate secrets, put temporary controls in place while the proper fix is prepared.

Eradication and recovery

Eradication means removing every foothold: malware, persistence mechanisms, rogue accounts, added keys, modified scheduled jobs. Most teams under-scope this. If you cleaned three machines and the attacker had a credential for a fourth, you cleaned nothing. Reset credentials broadly, including service accounts and API keys, not only the ones you saw misused.

Recovery is rebuilding from known-good sources and returning to service in a controlled order, with extra monitoring on restored systems for a while. Restoring from a backup taken after the compromise began just restores the compromise. Know your backup dates against your estimated intrusion start.

The phase everyone skips

Post-incident review. Within a couple of weeks, while memory is fresh, sit down and ask what happened, what worked, what did not, and what changes. Blameless does not mean consequence-free for process failures; it means you ask “why did the system allow this” before “who did this”. Put actions in a tracker with owners and dates. A report that nobody acts on is theatre.

The loop closes here: lessons feed back into preparation. New detection rules, a better runbook, an extra log source, a changed approval rule.

Communication is half the job

Technical teams underrate this. Leadership wants a short status at fixed intervals, even when the status is “no change”. Legal needs to know early if personal data may be involved, since statutory clocks may already be running (see the separate piece on CERT-In and the DPDP Act). Customer messages should be reviewed by people who know the facts and avoid guessing. Do not speculate about the attacker in writing until you can support it.

Keep incident chat channels and notes out of the compromised environment if you suspect the attacker can read your email. It has happened to well-resourced teams.

The lifecycle is a map. Real incidents jump backwards: you recover a server, find a new foothold, and drop back into analysis. That is normal. The loop in the diagram is not decoration.

Leave a Reply

Trending