What is alert fatigue in a SOC?
Alert fatigue is when analysts get so many low-value alerts that they skim or ignore them, real threats included. Causes, survey data, and fixes.
Definition
Alert fatigue
Also called Alarm fatigue, Alert overload, SOC alert fatigue
Alert fatigue is the desensitization that sets in when people receive more alerts than they can meaningfully review, most of them false or not worth acting on, so they start skimming, delaying, or ignoring alerts, including the real ones. The term comes from healthcare, where it describes clinicians tuning out monitor alarms and order-entry warnings. In a security operations center (SOC), it describes analysts facing a queue of noisy, duplicated, and context-poor detections.
- Average alerts a SOC team receives per day (Vectra, 2023)
- 4,484
- Time analysts spend manually triaging alerts each day (Vectra, 2023)
- 2.7 hours
- Share of the workday spent on incidents that aren't real threats (IBM, 2023)
- 32%
- Share of assigned alerts SOC staff get to in a typical day (IBM, 2023)
- 49%
How alert fatigue develops
Alert fatigue is a response to the ratio of alerts to real problems, not to volume alone. When most alerts turn out to be harmless, the people receiving them learn, reasonably, that the next one probably is too. They triage faster and less carefully, batch alerts for later, or close them on sight. The cost shows up on the day an alert is real and gets the same treatment.
The term comes from healthcare. The Agency for Healthcare Research and Quality's patient safety primer defines alert fatigue as busy workers becoming "desensitized to safety alerts," and cites a study in which monitors on 66 adult intensive care beds produced more than 2 million alerts in one month. The Joint Commission issued an alarm safety alert in 2013 and made alarm safety a National Patient Safety Goal in 2014.
Site reliability engineers describe the same loop for on-call pages. Google's SRE book warns that when pages come too often, engineers "second-guess, skim, or even ignore incoming alerts," sometimes missing a real page masked by the noise. Its rule is that every page should be actionable. A SOC alert queue fails the same test whenever most of what lands in it needs no action.
What causes alert fatigue in a SOC
Four causes account for most SOC alert fatigue, and they compound each other.
Noisy detections. Many rules fire on behavior that matches the pattern but is legitimate in your environment: an admin script, a backup job, an old software version the business knows about. An Oxford study of SOC analysts published at USENIX Security 2022 found that most of what analysts call false positives are "benign triggers," alerts that correctly match a rule but describe activity the organization has chosen to accept. The rule works as written; it just isn't worth a person's time. Vendors add to this. In Vectra's 2023 survey, 41% of analysts agreed that vendors flood them with pointless alerts because they are afraid of not flagging a breach.
Duplicate alerts. One phishing email can produce an email security alert, an endpoint alert when the attachment runs, an identity alert for the odd sign-in, and a SIEM correlation alert on top. Without grouping, each lands as a separate item for someone to open.
Missing context. An alert that names an IP address and a rule ID leaves the analyst to look up who owns the host, what the user normally does, and whether the indicator is known. The Oxford researchers found alarms were "lacking in the context needed by analysts to filter FPs from genuine alarms." In the IBM study, 81% of SOC staff said manual investigation slows their response.
Too many tools. Each console has its own queue, severity scale, and login. The SANS 2024 automation survey found that about 59% of organizations use more than 10 security tools in their SOC, and 76% of IBM's respondents said weak integration between security tools slows them down.
Why it matters for a SOC
Alert fatigue matters because the alerts nobody reviews are where intrusions hide. Two large surveys with published methods put numbers on the gap. Both are self-reported and commissioned by vendors that sell detection and response products, so read them as practitioner sentiment rather than measured telemetry.
Vectra commissioned Sapio Research to survey 2,000 security analysts at organizations with more than 1,000 employees across ten markets in March and April 2023. Respondents said their teams receive an average of 4,484 alerts a day and spend 2.7 hours a day triaging them by hand. They said they can't deal with 67% of daily alerts and that 83% of alerts are false positives. And 97% worried they would miss a relevant event buried in the flood.
Morning Consult surveyed 1,000 SOC team members for IBM between February 27 and March 18, 2023, 100 in each of ten countries, with a stated margin of error of plus or minus three points. Respondents said they spend 32% of a typical workday investigating incidents that aren't real threats, that 63% of the threats they review manually are low priority or false positives, and that they get to 49% of the alerts they're supposed to review.
The Oxford study is smaller (20 survey respondents and 21 interviews across seven SOCs) but goes deeper. One analyst put it plainly: "We know 99% of the alarms we generate are false positives, but we still have to look at them." The authors tie that validation work to "alarm burnout and eventually desensitization." Fatigue also shows up in the metrics a SOC reports: slower triage raises mean time to detect and respond, and a closed-without-review alert counts as handled.
How to reduce alert fatigue
Most fixes either cut how many alerts reach a person or raise what each one is worth. Teams usually need several.
- Tune detections. Track the disposition of every alert by rule. Rules that mostly produce benign triggers need an exception, a tighter condition, a lower severity, or retirement. Treat this as ongoing work, since every new tool and business change creates new benign patterns, and backtest a new rule for noise before it ships rather than tuning it afterward.
- Deduplicate and correlate. Group related alerts into one incident before anyone opens them. The Microsoft Defender portal correlates related alerts into incidents from Defender telemetry and from data ingested into Microsoft Sentinel, and Sentinel's automation rules can tag, assign, or close incidents as they're created or updated. Grouping only reaches alerts that flow into the platform doing it; an alert raised in a tool outside that platform still lands as its own item.
- Alert on risk, not on every event. Risk-based alerting in Splunk Enterprise Security writes low-confidence matches to a risk index as intermediate findings and creates a finding only when an entity's accumulated risk crosses a threshold, for example a score over 100 within 24 hours. The risk scores and thresholds are yours to assign and keep tuned, so the tuning work moves from individual alerts to scores rather than going away. Legacy SIEM vs modern SIEM covers how the newer designs handle alert volume instead.
- Enrich before triage. Attach asset owner, user history, threat intelligence, and prevalence to the alert automatically, so the analyst starts with the facts instead of six lookups. Enrichment also has to remember what the team already learned. Take an impossible-travel alert where the second sign-in comes from the corporate VPN: an analyst who doesn't know that address range checks it from scratch. Cotool keeps what its agents learn about your environment, such as "203.0.113.0/24 is corporate VPN egress," in shared memory files every agent reads and your team can inspect, so the next investigation starts from that fact.
- Automate the repeatable cases. Playbooks in a SOAR platform handle well-defined work end to end. SANS found phishing response is the most commonly automated process, at 52% of organizations. Playbooks only cover cases whose steps are known in advance.
- Have an agent investigate every alert. AI agents can do the first pass that playbooks can't script: query logs, enrich indicators, and reach a verdict with reasoning attached. The question to ask is which alerts an agent will take. Google Security Operations' Triage and Investigation Agent runs only on data ingested into Google SecOps SIEM, doesn't process alerts from the SOAR connector, and has no investigation queue, so alerts beyond its limit aren't analyzed automatically; within that scope it labels alerts as true or false positives with an explanation, typically in an average of 60 seconds. Microsoft's Phishing Triage Agent handles only user-reported phishing in Defender, needs Security Copilot capacity in Security Compute Units, and resolves false positives while leaving true positives open for an analyst. Cotool's agents take alerts from any connected source, whether or not Cotool stores the logs, and close or escalate each one with the evidence attached. What is an AI SOC covers how these platforms are built.
The first four shrink the queue. The last two change who works it.
Where an agent-native SOC differs
Most SOC tooling still assumes that alerts wait in a queue for a person, and treats fatigue as a queue-size problem. Agent-native platforms start from the assumption that every alert gets investigated, and that each verdict should feed back into the detection that raised it. Cotool, for example, has agents pick up every alert the moment it lands, gather context across connected tools, and close or escalate it with the evidence attached. For detections that run in Cotool, each false-positive or benign disposition feeds autotuning, which tests a change against the current version before applying it or sending it for review. That ties the investigation to step 1, the tuning step teams most often skip.
Detections that run in another SIEM can get the same loop if they live as code in Git. Ramp's rules run in two SIEMs, and for each false positive a Cotool agent asks why the rule fired on benign activity and opens a pull request against Ramp's detection repository with the proposed change and its reasoning. In Cotool's illustrative example, a rule on IAM role assumption from an unfamiliar IP fired 14 times in 7 days on CI runners after they moved to a new address range; the proposed diff excludes that role only from the new range, so the rule still fires for it from any other IP, and an engineer reviews and merges it (Cotool). Detections kept only in a SIEM's console still need their tuning done there.
Related terms
- False positive: an alert raised when no security event occurred. The Oxford study argues for separating these from benign triggers, which match a rule correctly but describe accepted activity.
- Alert triage: the first-pass decision on an alert: close, escalate, or investigate further.
- Risk-based alerting (RBA): alerting on accumulated risk per user or host instead of on each matching event.
- SOAR: security orchestration, automation, and response; platforms that run playbooks across security tools. See what SOAR is.
- AI SOC: a security operations model where AI agents investigate alerts. See what an AI SOC is.
FAQ
- What does alert fatigue mean?
- Alert fatigue means becoming desensitized to alerts because there are too many of them and most turn out to be unimportant. People respond more slowly, skim, or ignore alerts outright, and the real alert gets lost with the noise. The term is used in healthcare, site reliability engineering, and security operations.
- What is alert fatigue in cybersecurity?
- Alert fatigue in cybersecurity is what happens when SOC analysts receive more detections than they can investigate, so alerts go unreviewed or get closed without a real look. In a 2023 Morning Consult survey for IBM, SOC staff said they get to 49% of the alerts they're supposed to review in a typical day. Missed true positives are the main risk.
- What causes alert fatigue?
- Alert fatigue is caused by alert volume combined with low alert value. In a SOC the usual sources are detections that fire on benign activity, the same event alerting in several tools, alerts that arrive without the context needed to judge them, and too many separate consoles to check. Each one adds work without adding information.
- How do you reduce alert fatigue?
- You reduce alert fatigue by cutting the alerts a person has to look at and raising the value of the rest. Tune or retire rules that mostly fire on benign behavior, correlate duplicates into one incident, alert on accumulated risk per user or host, enrich alerts before they reach a person, and automate or delegate the first pass of triage to playbooks or AI agents.
- Is alert fatigue the same as alarm fatigue?
- Alert fatigue and alarm fatigue describe the same effect: people stop responding properly to warnings because most of them don't matter. Alarm fatigue is the usual term for audible bedside monitor alarms in hospitals, while alert fatigue covers software warnings, such as order-entry pop-ups or security detections. The fixes are similar in both fields: fewer, more meaningful alerts.
Sources
- 2023 State of Threat Detection - Vectra AI · accessed November 16, 2026
- Research Reveals Significant Disconnect Between Security Operations Teams and the Effectiveness of Threat Detection Tools - Vectra AI (GlobeNewswire) · accessed November 16, 2026
- Global Security Operations Center Study Results - Morning Consult for IBM · accessed November 16, 2026
- 99% False Positives: A Qualitative Study of SOC Analysts' Perspectives on Security Alarms - USENIX Security 2022 · accessed November 16, 2026
- 99% False Positives: A Qualitative Study of SOC Analysts' Perspectives on Security Alarms (paper) · accessed November 16, 2026
- The State of Automation in Security Operations: A SANS Survey · accessed November 16, 2026
- Alert Fatigue - AHRQ PSNet · accessed November 16, 2026
- Harm From Alarm Fatigue - AHRQ PSNet · accessed November 16, 2026
- Monitoring Distributed Systems - Google SRE Book · accessed November 16, 2026
- Analyze risk with risk-based alerting in Splunk Enterprise Security · accessed November 16, 2026
- Incidents and alerts in the Microsoft Defender portal · accessed November 16, 2026
- Automation in Microsoft Sentinel · accessed November 16, 2026
- Triage alerts with the Triage Agent - Google Security Operations · accessed November 16, 2026
- Microsoft Security Copilot Phishing Triage Agent in Microsoft Defender · accessed November 16, 2026
- AI SOC Without the Black Box - Cotool · accessed November 16, 2026
- How Ramp Scaled Detection & Response With Cotool · accessed November 16, 2026





