BLUE TEAM & INCIDENT RESPONSE
SOC operations · Logs · SIEM · Detection engineering · Threat hunting · Network & endpoint defence · Incident response · Forensics · Threat intel
The SOC and the Defender Mindset
- A SOC is a detection-and-response function, not a room full of screens: people, process and telemetry working together to shorten the time an attacker goes unnoticed.
- Tiered models (L1 triage, L2 investigation, L3 hunting and engineering) are still common, but mature SOCs blur the tiers and push automation into L1 work.
- MTTD and MTTR only mean something when you define the start and end timestamps precisely and measure them the same way every month.
- Alert fatigue is an engineering failure, not an analyst failure: every alert must be actionable, documented and owned, or it should be tuned or deleted.
- The defender mindset assumes breach, thinks in attack paths rather than single events, and treats every closed alert as data for improving detections.
A gated estate runs a 24-hour watchman team, and a SOC is the digital version of it. The watchmen sit in the guardroom with the camera feeds and the gate radios (the telemetry), they are trained people who can tell a resident from a thief (the analysts), and they follow an agreed drill — log it, call the supervisor, raise the alarm — so every shift reacts the same way (the processes). The famous "video wall" is just the guardroom screen; it is the trained eyes and the shift drill, not the screen, that actually keep the estate safe.
► What a Security Operations Centre Actually Does
Every offensive module in GHOSTNET ends in the same place: an attacker on a network doing something they should not. This module is about the other side of that story. A Security Operations Centre (SOC) is the function inside an organisation, or provided to it by a managed service, that watches for those attackers, decides which signals are real, and drives the response when they are. It is tempting to picture a dark room with a video wall, but the room is the least important part. A SOC is three things working together: telemetry (the logs and sensor data you collect), people who can interpret that telemetry, and processes that turn interpretation into consistent action.
The core loop is simple to state and hard to do well. Collect data from endpoints, servers, network sensors, identity providers and cloud platforms. Run detections against that data. Triage the alerts those detections produce. Investigate the ones that look real. Respond by containing and eradicating the threat. Then feed what you learned back into the detections and the hardening work so the same attack is caught earlier, or blocked outright, next time. Everything else in this module, from Sigma rules to Volatility, is a tool for one stage of that loop.
► Tiers and Roles
The classic SOC structure splits analysts into tiers by depth of investigation. The model is still widely used for staffing and career paths, even though modern teams increasingly organise around functions (detection engineering, incident response, threat hunting) rather than strict escalation ladders.
| Role | Main Responsibility | Typical Output |
|---|---|---|
| L1 / Triage analyst | Work the alert queue, enrich, decide benign vs suspicious, escalate | Closed alerts with notes, escalated tickets |
| L2 / Incident responder | Deep investigation, scoping, containment actions, coordination | Incident timeline, containment decisions, IR ticket |
| L3 / Threat hunter | Proactive hypothesis-driven searches for undetected activity | Hunt reports, new detection ideas |
| Detection engineer | Write, test, tune and version detections; own detection coverage | Rules in a repo, coverage maps, tuning changes |
| Threat intelligence analyst | Track adversaries, produce relevant intel, manage IOCs/TTPs | Intel reports, curated feeds, hunt leads |
| Forensic analyst | Evidence acquisition and analysis of disks, memory and logs | Forensic reports, artefact timelines |
| SOC manager / IR lead | Staffing, metrics, stakeholder communication, incident command | Metrics reports, post-incident reviews |
In a small company one person may wear all of these hats. In a large enterprise or MSSP each may be a team. What matters is that every responsibility has an owner. A detection that nobody owns will decay: the log source changes format, the rule silently stops matching, and nobody notices until the incident review.
► The Alert Lifecycle
An alert moves through predictable states. Understanding them is how you avoid the two classic failures: closing a real incident as noise, and drowning a real incident in a queue of noise.
Notice the category "benign true positive". The rule fired correctly, the behaviour really happened, but it was authorised: an administrator ran PsExec during a maintenance window, or a vulnerability scanner triggered a port scan alert. Separating these from false positives (the rule matched something it should not have) matters, because the fix is different. A false positive means the rule logic is wrong. A benign true positive means you need context, such as an allowlist of scanner IPs or a change-ticket lookup.
► Metrics: MTTD, MTTR and Their Traps
Two metrics dominate SOC reporting. Mean Time To Detect (MTTD) is the average time from when malicious activity begins to when the SOC becomes aware of it. Mean Time To Respond (MTTR, sometimes split into Mean Time To Contain and Mean Time To Recover) is the average time from detection to the threat being contained or resolved. Industry reports such as Mandiant M-Trends track global median dwell time (start of intrusion to detection), which has fallen from over 400 days a decade ago to roughly 10-11 days in recent years, while ransomware operators can go from initial access to encryption in under 24 hours. The gap between those numbers is the whole argument for investing in detection.
| Metric | Start timestamp | End timestamp | Common trap |
|---|---|---|---|
| MTTD | First malicious event (often known only after investigation) | First alert or human awareness | Only measurable after the fact; easy to fudge by picking a late start |
| MTTA (acknowledge) | Alert created | Analyst picks it up | Gamed by clicking acknowledge without working the alert |
| MTTC (contain) | Detection | Host isolated / account disabled | Ignores partial containment that missed other hosts |
| MTTR (resolve) | Detection | Incident closed | Pressure to close early hides incomplete eradication |
| False positive rate | n/a | n/a | Per-rule rate matters more than the global average |
Good SOCs also measure things that show coverage and quality rather than speed: the percentage of critical assets sending logs, the number of ATT&CK techniques with at least one tested detection, the share of alerts that are auto-enriched, and how many incidents were found by a hunt or a third party rather than by your own alerts. A rising share of incidents reported by outsiders is a red flag that detection is failing.
► Alert Fatigue
Alert fatigue is what happens when analysts see so many low-value alerts that they stop investigating carefully. It is the single most common reason real intrusions get closed as noise; post-incident reviews regularly find that the critical alert fired, was looked at for thirty seconds, and was closed because the same rule had produced 400 false positives that month. The fix is engineering discipline, not more staff:
| Practice | What it means in practice |
|---|---|
| Every alert is actionable | If the analyst has no decision to make, it is a report or a dashboard, not an alert |
| Every rule has a runbook | What to check, what benign looks like, when to escalate |
| Every rule has an owner | Someone reviews its false positive rate monthly |
| Severity reflects impact | High means drop everything; if everything is high, nothing is |
| Automate enrichment | SOAR adds user, asset, geo-IP, reputation and recent alerts before a human looks |
| Risk-based alerting | Low-fidelity signals add risk scores to an entity; alert only when the score crosses a threshold |
| Delete dead rules | A rule that has never fired true in a year, and is not covering a critical technique, is cost without value |
Risk-based alerting deserves emphasis. Instead of firing an alert every time someone runs whoami, a rule adds 10 risk points to that user and host. Encoded PowerShell adds 40. A new service install adds 30. When an entity crosses 100 within 24 hours, a single high-fidelity alert fires that lists all the contributing observations. Splunk Enterprise Security, Microsoft Sentinel (via entity behaviour analytics and incident grouping) and Elastic all support variants of this approach.
► AI in the 2026 SOC
By 2026 most SIEM and EDR vendors ship LLM-based assistants that summarise alerts, draft queries and propose triage verdicts, and many teams run agentic workflows that enrich and pre-triage L1 alerts automatically. These tools genuinely cut toil, but they inherit the quality of the underlying telemetry and they can be confidently wrong. Treat their output as a junior analyst's notes: useful, fast, and always verified against the raw events before an action like disabling an executive's account. Attackers also target these pipelines, for example by planting prompt-injection text in fields an AI assistant will read, such as email subjects or user-agent strings.
► The Defender Mindset
John Lambert's well-known line is that defenders think in lists while attackers think in graphs. A defender with a list sees ten unrelated medium alerts. An attacker sees a path: phishing email, user workstation, cached admin credential, file server, domain controller. Thinking like a defender means deliberately thinking in graphs too. When you see one suspicious event, ask what must have happened before it for it to be possible, and what the attacker would do next if it is real. Then go and look for those events.
1. What exactly happened? (process, user, host, time, command line) 2. Is it expected here? (baseline: has this host/user done this before?) 3. What happened just before it? (parent process, logon, email, download) 4. What happened just after it? (network connections, child processes, new files) 5. Where else has it happened? (same hash, same domain, same user on other hosts) 6. What is the blast radius? (privileges of the account, data on the host) 7. What do I do now? (close with evidence, escalate, or contain)
Finally, assume breach. Prevention will fail eventually: a zero-day, a phished MFA push, a contractor's laptop. A defender who assumes breach invests in visibility and response, practises incident handling before it is needed, and measures success not by "nothing happened" but by how quickly and completely the team found and removed the thing that did.
- A SOC is a detection-and-response function, not a room full of screens: people, process and telemetry working together to shorten the time an attacker goes unnoticed.
- Tiered models (L1 triage, L2 investigation, L3 hunting and engineering) are still common, but mature SOCs blur the tiers and push automation into L1 work.
- MTTD and MTTR only mean something when you define the start and end timestamps precisely and measure them the same way every month.
- Alert fatigue is an engineering failure, not an analyst failure: every alert must be actionable, documented and owned, or it should be tuned or deleted.
- The defender mindset assumes breach, thinks in attack paths rather than single events, and treats every closed alert as data for improving detections.