Skip to main content

SOC False Positives: How Analysts Triage Benign Triggers

Separate false alarms from benign triggers, run a seven-step triage worksheet, and practice with a labeled fictional privileged VPN alert.

Updated by

JW
Jack WalshNov 14, 2026 · 8 min read

Hi, I'm Jack, the owner of Cybersecurity Jobs List and co-founder of Himalayas (himalayas.app) and Cavuno (cavuno.com). Across all my platforms, I work with application security daily: dependency vulnerability scanning, secure authentication, API security, and data protection across hundreds of thousands of users. My technical background is in computer science (UNSW). I studied security engineering and computer networks, and worked as a research assistant on VR experiments that were published in the Journal of Experimental Psychology. I also work with cybersecurity hiring data every day, tracking which companies are posting, what certifications actually appear in listings, how salaries differ by sub-discipline and clearance level, and where the talent gaps are widest. That combination of security practice, engineering at scale, and daily immersion in hiring data shapes the content on this site. I'm currently based in Sydney, Australia.

Share this post

Abstract illustration of noisy alerts sorted into three bins beside a checklist panel.

Most people who say "false positive" mean "this alert wasted my time." That everyday use hides three different outcomes: a bad match, a correct match that is still safe business activity, and a real attack you closed too early. If you cannot tell those apart, you will tune the wrong thing, escalate the wrong thing, or freeze when an interviewer asks how you handle noise.

This guide is for SOC analysts and early detection staff who need a practical triage method. It is not a product pitch and it does not invent a universal false-positive percentage for every SOC.

False alarm, benign trigger, and true positive

A useful starting point comes from research with SOC practitioners, including the USENIX Security 2022 paper 99% False Positives (Alahmadi, Axon, and Martinovic). Analysts in that work often used "false positive" loosely, then drew a sharper line once they talked through real cases:

OutcomeWhat matchedWhat it means for youTypical next step
False alarmLogic fired without a real security-relevant event (bad signature, broken enrichment, wrong field)The detection content is wrong or incompleteDocument the defect; send a clear feedback note to the rule owner
Benign triggerLogic correctly matched real activity that has a legitimate business explanationThe event is real; the risk is not (today)Record the explanation, evidence, and whether an exception should be permanent, temporary, or refused
True positive / suspiciousActivity that is malicious, unauthorized, or still unresolved after reasonable checksYou still own investigation qualityEscalate or contain under policy; keep unknowns explicit

That split matters. Retiring a rule that only ever fires on approved scanners is different from silencing a rule because travel logins are common. One is detection quality. The other is exception design. The paper also notes that validation work is tedious and can contribute to burnout when analysts must re-prove the same benign patterns without better context or explainability.

Do not treat any single survey quote as your organization's measured false-positive rate. Measure disposition rates on your own top alert types instead.

Where this fits in incident response

Triage is the front door of detection work. NIST's current incident response guidance, SP 800-61 Revision 3 (April 2025), places Detect, Respond, and Recover inside broader cybersecurity risk management aligned to the Cybersecurity Framework 2.0. Preparation work in Govern, Identify, and Protect still shapes what you see, but the alert on your screen is a Detect problem first: can you decide quickly and defensibly what happened?

If your notes skip evidence and jump to "looks fine," you are not only creating audit pain. You are hiding the feedback detection engineers need to improve content quality. For ownership differences between analysts and detection engineers, see detection engineer vs SOC analyst.

A practical triage worksheet

Copy this into your case system or a personal template. Answer in short sentences. Prefer observed facts over vibe.

  1. Alert identity. Rule or detection name, severity, first/last seen in this case, affected user/host/app if named.
  2. What the rule claims. One sentence: which behavior is supposed to be bad, and which log fields prove that claim.
  3. Evidence gathered. List sources you actually opened (EDR process tree, identity logs, email gateway, ticketing, change calendar). Do not list tools you did not check.
  4. Benign explanations tested. Travel, approved scanner, service account job, change window, new hire onboarding, known automation. Mark each as supported, contradicted, or unknown.
  5. Risk leftovers. Privilege level, data sensitivity, unusual pairing (new device + impossible travel + disabled MFA), lateral movement signs, persistence signs.
  6. Decision and confidence. False alarm / benign trigger / suspicious / incident. State confidence as high, medium, or low and name the missing evidence that would change your mind.
  7. Feedback or escalation. If the rule is noisy or opaque, write a feedback note. If policy requires handoff, escalate with a clean summary, not a paste dump.

Strong case notes separate evidence, assessment, and unknowns. The same habit shows up in good incident write-ups; see the labeled fictional packet in SOC analyst incident report example.

Fictional example: night-shift VPN spike

This case is labeled fiction for training. It is not a real employer incident and not advice to close production alerts without your playbooks.

Alert. Medium severity: "Successful VPN auth from rare ASN for privileged user." User: j.lee (IT admin). Time: 02:14 local. Source IP geolocates to a country the user has never used in the last 90 days of available logs.

What the rule claims. Privileged VPN success from a rarely seen network is worth human review because valid accounts are a common abuse path. Mapping candidates may include MITRE ATT&CK Valid Accounts (T1078), but mapping is a hypothesis aid, not proof. For disciplined mapping language, use the ATT&CK mapping guide.

Evidence gathered (fictional).

  • Identity provider: success, MFA satisfied with hardware key, device certificate present.
  • EDR on the admin workstation: no concurrent interactive logon; host is locked at home per last check-in.
  • HR/travel tool: approved emergency travel started yesterday; ticket includes the same country.
  • Change calendar: no privileged maintenance window named for this user.
  • VPN community string / posture: corporate-managed laptop posture healthy.

Benign explanations. Approved travel is supported. Impossible travel against a second session is not present. MFA bypass is not present in the available fields.

Risk leftovers. The account is privileged, so a stolen session still matters. There is no second factor failure, no impossible concurrent session, and no endpoint malware signal in the checks above. Residual risk is low-to-medium until the user confirms the session in the morning if policy requires it.

Decision. Benign trigger with medium confidence pending optional user confirm per local policy. Not a false alarm: the rare ASN match was real. Not a closed incident: you did not invent certainty beyond the logs.

Feedback note to detection (example). "Rule X correctly fires on rare ASN for privileged VPN. In the last seven days this queue has three similar travel cases. Please add enrichment for open travel tickets and device posture before page, or split severity when MFA + managed device + travel ticket are all true. Happy to retest."

That note is gold for detection engineering. It names the rule, the pattern, the desired change, and a retest offer. It does not demand a silent suppress-all exception without evidence.

What good alarms need from your side

The same USENIX work summarizes qualities that make alarms faster to validate. In plain language, analysts need alarms that are:

  • Reliable enough that trust is earned, not assumed
  • Explainable so you can see why it fired
  • Analytical enough to point at useful next evidence
  • Contextual about users, assets, and business process
  • Transferable so the next analyst can continue your case

You cannot rebuild the SIEM on night shift. You can still refuse to close with empty notes, refuse to accept black-box "AI said benign" without evidence, and refuse to leave the same broken rule undocumented for the next three people.

Common failure modes

  • Closing as false positive when you mean benign trigger. Detection owners then "fix" the wrong problem.
  • Exception sprawl. Permanent allowlists without owners, expiry, or scope become attacker cover.
  • Tool worship. Opening every console without a question list wastes the queue and still misses the deciding field.
  • Confidence theater. Writing "confirmed malicious" when you only have a weak correlation.
  • No feedback loop. The noisiest rules survive because nobody files a reproducible note.

Interview and portfolio use

Interviewers often ask how you handle alert fatigue. Weak answers complain about volume. Strong answers show method:

  1. Explain false alarm vs benign trigger in one minute.
  2. Walk a labeled fictional case with evidence, unknowns, and decision.
  3. Show a feedback note you would send to detection.
  4. State what you would measure for one noisy rule (disposition mix for two weeks, not a global FP myth).

Practice with behavioral structure if needed: STAR examples for SOC work and the broader SOC interview questions guide. Keep confidential employer data out of public portfolios.

One-week practice plan

  1. Day 1. Pick five public technique pages and write the behavior in plain language before any tool talk.
  2. Day 2. Build the seven-step worksheet in your notes app. Run two fictional alerts end to end.
  3. Day 3. Rewrite one weak case note you have seen (or invent a weak one) into evidence / assessment / unknowns.
  4. Day 4. Draft three detection feedback notes for noisy patterns (scanner, MFA travel, service account batch job).
  5. Day 5. Time-box a 15-minute triage drill. Stop when the clock ends; list what you still do not know.
  6. Day 6. Map one case to a single ATT&CK technique only if evidence supports it; otherwise write "insufficient for mapping."
  7. Day 7. Do a mock interview answer out loud. Record yourself once. Cut filler and unsupported rates.

None of this requires attacking systems you do not own.

What to do next on the job board

When you read SOC postings, look for clues about triage quality expectations: mention of playbooks, case quality metrics, detection feedback, or only raw ticket counts. Browse live roles on the job board, including SOC analyst filters when you want operations seats, and compare shift language against engineering-heavy detection roles.

Clear triage is a career skill, not a complaint about vendors. If you can name the outcome type, show your evidence, and leave a useful note for the next human, you are already ahead of volume-only noise handling.

Related posts

We use cookies

We use cookies to understand how this board is used and to improve it. Analytics run only if you accept. Cookie Policy