Agent Foskett Academy • SOC Analyst Academy • Module 1 • Lesson 2 • Inside the SOC

Lesson 2 — The Alert Says “Suspicious” — What Does That Actually Mean?

You have selected the first alert from the queue. Now comes the next SOC mistake to avoid: treating the alert title as the conclusion.

“Suspicious PowerShell activity.” “Unusual sign-in.” “Potential credential theft.” Those phrases are detection claims. They tell you why a security product wants your attention. They do not, by themselves, prove what happened.

An alert is the beginning of the investigation — not the end of it.
Agent Foskett SOC Analyst Academy understanding security alerts
The alert in front of you

Microsoft Defender has raised a high-severity alert for suspicious PowerShell activity on a finance laptop. Your job is to work out exactly what the detection says — and what it does not say.

✓ Read the alert title as a claim
✓ Identify the triggering evidence
✓ Separate facts from interpretation
✓ Define what still needs to be proven

Case briefing

07:11 — FIRST ALERT OPENED SEVERITY High ALERT TITLE Suspicious PowerShell activity DEVICE Finance-LT-044 USER alex.w@contoso.com OBSERVED powershell.exe executed with an encoded command ALERT DESCRIPTION PowerShell activity associated with suspicious command-line behaviour was detected. THE TEMPTATION "PowerShell malware ran on the finance laptop." THE PROBLEM The alert did not actually prove that.

Investigation objective

Learn how to read a security alert as an analyst. Identify the detection claim, triggering evidence, affected entities, timestamps and supporting signals before deciding whether the activity is malicious, benign or still unexplained.

Investigator's rule

Do not promote detection language into fact. “Suspicious” means the observed activity matched logic that deserves investigation. It does not automatically mean compromise occurred.

Stage 1 — translate the alert into plain English

Before pivoting anywhere, rewrite the alert in neutral language. Strip away words such as suspicious, malicious, anomalous and compromised until only the observable claim remains.

Alert language Neutral analyst translation
Suspicious PowerShell activity PowerShell executed behaviour that matched this detection's suspicious criteria.
Impossible travel Authentication events were observed from locations that appear inconsistent within the measured time window.
Potential credential theft Activity associated with known credential-access techniques was observed.
Unusual sign-in The sign-in differed from an established or expected pattern.

Why neutral language matters

If you begin with “malware ran,” every later event risks being interpreted to support that theory. If you begin with “PowerShell executed an encoded command,” you can investigate what the command did without deciding the answer first.

The detection has a point of view

Every alert is produced by logic: analytics, behavioural models, threat intelligence, correlations or product detections. Understanding what triggered that logic helps you understand how much the alert actually tells you.

Stage 2 — separate the alert into six questions

READ THE ALERT 1. WHAT? What behaviour triggered the detection? 2. WHO? Which user, account or identity is involved? 3. WHERE? Which device, mailbox, workload or cloud resource? 4. WHEN? What is the actual event time? 5. WHY? Why did the product consider the behaviour suspicious? 6. WHAT NEXT? What must the analyst verify before reaching a conclusion?

Entities give you pivots

The user, device, IP address, process, URL, mailbox or cloud resource named in an alert are not decoration. They are the starting points for the next investigation steps.

Timestamps give you boundaries

The alert creation time may not be the same as the event time. Analysts need the timestamp of the underlying activity so they can examine what happened immediately before and after it.

Stage 3 — identify fact, interpretation and unknown

This is one of the most useful habits in SOC work. For every important statement, decide which of three buckets it belongs in.

Category Example from this alert Analyst treatment
Fact powershell.exe executed on Finance-LT-044. Supported directly by telemetry.
Fact The command line contained encoded content. Supported directly by the observed command line.
Interpretation The encoded command was malicious. Requires investigation.
Unknown What process launched PowerShell? Pivot to process telemetry.
Unknown What did the encoded command actually do? Decode or inspect safely and correlate behaviour.
Unknown Did PowerShell create files or make network connections? Check related endpoint telemetry.

Encoded does not equal malicious

Attackers use encoded PowerShell, but administrators and legitimate software can too. Encoding increases investigative interest; it does not remove the need to examine the command and surrounding behaviour.

Suspicious does not equal false positive either

The opposite mistake is dismissing the alert because legitimate explanations exist. Your task is not to defend or condemn the activity. Your task is to determine which explanation the evidence supports.

Stage 4 — ask what evidence would change your mind

Good analysts do not merely collect more telemetry. They look for evidence that can discriminate between competing explanations.

WORKING EXPLANATIONS A. Legitimate administration B. Approved software activity C. User-initiated script D. Malicious execution WHAT WOULD HELP? Parent process ↓ Full command line ↓ Script / decoded content ↓ File creation ↓ Network connections ↓ Persistence activity ↓ Other alerts on user or device ↓ Known change / admin context

Look for corroboration

A single suspicious command is one signal. A suspicious command followed by a dropped executable, outbound connection and persistence mechanism is a very different evidentiary picture.

Look for contradiction

Evidence can also weaken the malicious hypothesis. A known management process launching an approved signed script during a documented maintenance window may explain the detection.

Stage 5 — define the investigation before you query

Agent Foskett does not begin with a giant hunt across every table. The alert has already given us a focused set of questions.

INVESTIGATION QUESTIONS 1. What launched powershell.exe? 2. What was the complete command line? 3. What did the encoded content contain? 4. Did PowerShell create or modify files? 5. Did it connect to an external destination? 6. Did it establish persistence? 7. Is alex.w associated with the execution? 8. Are there related alerts on Finance-LT-044? 9. Is there an approved administrative explanation? 10. Does the combined evidence support malicious activity?
Do not ask the logs to prove the alert. Ask the logs to explain the activity.

Agent Foskett's alert-reading workflow

OPEN ALERT ↓ READ TITLE ↓ READ DESCRIPTION ↓ IDENTIFY TRIGGERING BEHAVIOUR ↓ IDENTIFY ENTITIES User • Device • IP • Process • Resource ↓ CHECK EVENT TIME ↓ SEPARATE Facts • Interpretation • Unknowns ↓ WRITE INVESTIGATION QUESTIONS ↓ PIVOT TO EVIDENCE ↓ UPDATE THE THEORY

Your alert-reading evidence board

Alert element What to ask Common mistake
Title What is the detection claiming? Treating the title as proven fact.
Severity How strongly has the product prioritised it? Assuming severity proves impact.
Description What behaviour or logic triggered the alert? Skipping directly to the entities.
Entities What user, device, IP, process or resource can I pivot from? Investigating the alert without expanding context.
Timestamp When did the underlying activity occur? Using alert creation time as the event time.
Evidence Which observations are directly supported? Mixing interpretation with telemetry.

Write the initial assessment like an analyst

Example: Microsoft Defender generated a high-severity alert after PowerShell executed an encoded command on Finance-LT-044 in the context of alex.w. At this stage, telemetry confirms PowerShell execution and encoded command-line content, but the alert alone does not establish that the activity was malicious. Investigation should determine the parent process, command contents, subsequent file and network activity, related alerts and whether an approved administrative explanation exists before assigning a final disposition.

Lesson 2 key takeaways

  • An alert title is a detection claim, not an investigation conclusion.
  • Translate loaded alert language into neutral, observable behaviour.
  • Identify what triggered the detection before chasing every related event.
  • Extract the user, device, process, IP address, resource and event time as investigation pivots.
  • Separate facts, interpretations and unknowns.
  • “Suspicious,” “unusual” and “potential” all require validation.
  • Look for evidence that can both support and contradict your working theory.
  • Correlated behaviour is usually more informative than one isolated signal.
  • Write investigation questions before writing a conclusion.
  • The purpose of the investigation is to explain the activity — not to prove the alert correct.

Module 1 — inside the SOC

You now understand what the alert is actually claiming. The next challenge is deciding whether the severity assigned by the product should control your response.

Next: Lesson 3 — High Severity Does Not Always Mean High Priority

Continue your SOC Analyst training

Module 1 builds the analyst mindset: triage, severity, evidence, context and defensible decisions before deep investigation begins.

How to read a security alert in a SOC

Lesson 2 of the Agent Foskett SOC Analyst Academy teaches analysts how to understand security alert titles, descriptions, entities, timestamps and triggering evidence before beginning a deeper investigation.

Security alert investigation and evidence validation

Learn how to separate detection claims from facts, identify investigation unknowns, build evidence-driven questions and avoid treating words such as suspicious, unusual or potential as proof of malicious activity.