TTechclick ⚡ XP 0% All lessons
CrowdStrike · Endpoint Security · Detection EngineInteractive · L1 / L2 / L3

CrowdStrike Falcon NGAV & EDR — IOAs vs IOCs, Machine Learning & How Detections Work

Falcon's real power is how it decides what is malicious. This lesson opens the detection engine: NGAV with on-sensor and cloud machine learning, Indicators of Attack (IOAs) versus Indicators of Compromise (IOCs), behavioural prevention and exploit mitigation, EDR's continuous recording, and exactly how a single suspicious action is scored and surfaced as a detection — so you can explain 'why behaviour beats signatures' in any interview.

📅 2026-06-19 · ⏱ 16 min · 5 infographics · live detection demo · 🏷 10-Q assessment + AI Tutor inline

⚡ Quick Answer

A clear, interactive guide to the CrowdStrike Falcon detection engine (2026): NGAV with on-sensor and cloud machine learning, Indicators of Attack (IOAs) vs Indicators of Compromise (IOCs), behavioural prevention, EDR continuous recording and detections, prevention policy settings, sensor visibility and quarantine, exploit mitigation, and exactly how a detection is scored and surfaced.

🎯 By the end you will be able to

Read as:

Pick where you want to start

1

IOA vs IOC

Behaviour and intent vs leftover forensic artifacts.

2

NGAV & machine learning

Known IOCs, on-sensor ML and cloud ML.

3

Behaviour & EDR

IOAs, exploit mitigation and continuous recording.

4

Scoring & policy

How a detection is scored, and how to tune policy.

🧠 Warm-up — 3 questions, no score

Just notice which ones make you pause. We answer all three inside the lesson.

1. Which one describes attacker behaviour, not a leftover artifact?

Answered in IOA vs IOC.

2. What catches brand-new malware that has no known hash?

Answered in NGAV & machine learning.

3. What does Falcon EDR do that NGAV prevention alone does not?

Answered in Behaviour & EDR.

Most engineers think…

Most people think endpoint detection is 'a list of bad file hashes the antivirus checks against'. That mental model is exactly what modern attackers defeat for breakfast.

Falcon's detection engine is behaviour-first. It still blocks known-bad with IOCs (hashes, IPs, domains), but its edge is Indicators of Attack (IOAs) — the chain of actions an attack must perform (execute, hide, persist, reach out) — plus on-sensor and cloud machine learning for unknown malware, exploit mitigation for memory tricks, and EDR continuous recording that catches anything prevention misses. Attackers can repack a file in seconds, but they cannot avoid the behaviours the attack requires. Understanding that is what lets you explain 'why behaviour beats signatures' in an interview.

① IOA vs IOC — behaviour and intent vs leftover clues

Start with the one comparison every interviewer loves. An IOC is a clue an attacker left behind — a file hash, a malicious IP or domain, a registry key. It is evidence the security of a system was breached. The problem: it is reactive. By the time you have the hash, the damage may be done, and the attacker only has to repack the file to get a brand-new hash.

An IOA is different: it describes what the attacker is trying to do, regardless of the malware or exploit used. A successful attack must perform a chain of actions — run code, hide in memory, gain persistence across reboots, call out to a command-and-control server. IOAs watch for that behaviour in real time.

Why behaviour beats signatures

CrowdStrike's classic line: IOCs are like a witness describing a bank robber's purple van and cap — useless the moment he switches to a red car and a crowbar. IOAs catch the robbery itself. Attackers change files constantly, but they cannot avoid the actions an attack requires. That is why a behaviour-first engine catches fileless and never-seen-before threats that pure signature matching misses entirely.

Figure 1 — IOA vs IOC at a glance
IOCs are leftover artifacts you find after the fact; IOAs catch the attacker's behaviour as it happens.IOA vs IOC at a glanceIOC (artifact)File hash, IP, domainFound after the factEasy to change / repackReactive — what happenedIOA (behaviour)Chain of attacker actionsCaught in real timeHard to avoid — attack needs itProactive — what's happening
IOCs are leftover artifacts you find after the fact; IOAs catch the attacker's behaviour as it happens.
Quick check · Q1 of 10 · Understand

Which statement best captures the IOA vs IOC difference?

Correct: c. An IOA describes what the attacker is doing — the chain of actions an attack must perform — so it catches the attack regardless of the file. An IOC (hash, IP, domain) is a forensic artifact left behind, useful only after the fact and easy to change.
👉 So far: IOA = attacker behaviour and intent (the action chain), caught in real time and hard to avoid. IOC = leftover artifact (hash/IP/domain), found after the fact and easy to change. Behaviour beats signatures.

② NGAV & machine learning — blocking known and unknown malware

NGAV (Falcon Prevent) is the prevention front line, and it works in layers. The fast, cheap layer is known-malware blocking: matching file hashes and other IOCs against CrowdStrike intelligence and your own custom IOC lists. If it is known-bad, it never runs.

The real upgrade is machine learning for malware nobody has seen before. Falcon runs ML in two places. On-sensor ML scores a file's maliciousness locally, so it can block a brand-new binary even when the endpoint is offline. Cloud ML, trained on the rich telemetry of the Security Cloud, adds a deeper verdict and continuously improves without a fleet-wide reinstall.

Sensitivity is a dial, not a switch

The ML engines have sensitivity levels — typically Disabled, Cautious, Moderate, Aggressive and Extra Aggressive. Higher levels catch more but risk more false positives. Crucially, you set the level separately for detection and for prevention: you can let Falcon detect aggressively while it only blocks at a more conservative level until you trust it.

Figure 2 — The NGAV detection layers
Falcon Prevent stacks fast known-bad blocking, machine learning and behaviour — cheapest and surest at the top.The NGAV detection layersKnown IOCsHashes / IPs / domains blockedOn-sensor MLScores unknown files offlineCloud MLDeeper verdict from Security CloudBehavioural IOAsBlocks the attack chainExploit mitigationStops memory-based tricks
Falcon Prevent stacks fast known-bad blocking, machine learning and behaviour — cheapest and surest at the top.
🎯
Indicator of Attack (IOA)
tap to flip

Describes attacker behaviour and intent — the chain of actions an attack must perform. Catches an attack regardless of the file used, in real time.

🔍
Indicator of Compromise (IOC)
tap to flip

A forensic artifact left behind — a hash, IP or domain. Useful after the fact, but easy for an attacker to change by repacking the file.

🧠
On-sensor + cloud ML
tap to flip

On-sensor ML scores unknown files locally (even offline); cloud ML adds a deeper verdict trained on the Security Cloud. Sensitivity is a dial set per policy.

📼
EDR continuous recording
tap to flip

Falcon Insight records endpoint activity like a DVR, so threats that slip past prevention are still detected and can be replayed step by step.

Say 'attackers change files, not behaviour'

In an interview, this one line nails IOA vs IOC: an attacker can repack a file to defeat any hash/IOC in seconds, but cannot avoid the actions an attack requires — execute, hide, persist, call out. IOAs watch those actions, which is why behaviour-first detection beats signatures on fileless and never-seen threats.

Quick check · Q2 of 10 · Understand

How does Falcon NGAV catch brand-new malware that has no known hash?

Correct: b. Known-bad is blocked by IOC/hash matching, but unknown malware is caught by machine learning: on-sensor ML scores files locally (even offline) and cloud ML adds a deeper verdict trained on the Security Cloud.
👉 So far: NGAV layers: known-IOC/hash blocking, on-sensor ML (scores unknown files offline), and cloud ML (deeper verdict from the Security Cloud). ML sensitivity is a dial set separately for detection and prevention.

③ Behaviour & EDR — IOAs, exploit mitigation and continuous recording

On top of ML sits the behavioural layer. Behavioural IOAs watch sequences of actions — a document spawning PowerShell that downloads and runs a payload, a process injecting into another, an attempt to disable defences — and block the chain even when every individual file looks clean. AI-powered IOAs extend this: cloud ML analyses events at runtime, generates new behavioural detections and pushes them to the sensor, which correlates them with local events to decide maliciousness. Exploit mitigation covers memory-based tricks like buffer overflows and code injection that never drop a file at all.

Prevention is never perfect, so Falcon also records. EDR (Falcon Insight) acts like a DVR on the endpoint, capturing process starts, network connections, file writes and registry changes.

Why the recorder matters

Because it records everything, EDR catches the threat that slipped past prevention: applying behavioural analytics over the stream, it raises a detection and lets an analyst replay exactly what happened — who, what, in what order — instead of guessing from a single quarantined file.

Figure 3 — AI-powered IOAs — cloud to sensor
Cloud ML analyses runtime events, generates new behavioural detections and pushes them to the sensor to act.AI-powered IOAs — cloud to sensorObservesensor watches actionsStreamevents to SecurityCloudGeneratecloud ML makes IOAsPushIOAs sent to sensorActcorrelate + blockchain
Cloud ML analyses runtime events, generates new behavioural detections and pushes them to the sensor to act.
Figure 4 — One detection engine, many signals
Falcon fuses several detection methods into one scored verdict for each event.One detection engine, many signalsDetection enginescores + severityKnown IOCsOn-sensor MLCloud MLBehavioural IOAsExploit mitigationEDR recording
Falcon fuses several detection methods into one scored verdict for each event.
'NGAV blocked it, so EDR is optional' under-sell

Prevention is never 100 percent. The reason EDR records everything like a DVR is precisely to catch what slips past NGAV. Treating EDR as optional means a fileless or living-off-the-land attack that dodges prevention leaves no trail to investigate. Always pair prevention with continuous recording.

▶ Watch a fileless attack get caught by behaviour, not signature

How a clean-looking document becomes a blocked detection. Press Play for the healthy path, then Break it to see the classic failure.

① ActionA user opens a document that quietly spawns PowerShell to download and run a payload — no known-bad file is involved.
② IOA matchBehavioural IOAs recognise the action chain (document ▸ PowerShell ▸ download ▸ execute) as an attack pattern, not a benign macro.
③ ScoreThe engine scores the chain high-confidence, assigns a severity and correlates it with cloud-generated AI IOAs for the verdict.
④ Prevent + detectPrevention kills the chain and quarantines the payload; a detection is raised in the console with the full process tree and ATT&CK technique.
Press Play to step through the healthy behavioural-detection path. Then press Break it.
Quick check · Q3 of 10 · Apply

A clean-looking document spawns PowerShell that downloads and runs a payload, with no known-bad file involved. What stops it?

Correct: a. No single file is known-bad, so signatures and hashes miss it. Behavioural IOAs watch the sequence of actions — document spawns PowerShell, downloads, executes — and block the chain itself, which is the whole point of behaviour-based detection.
👉 So far: Behavioural IOAs block the attack chain; AI-powered IOAs are generated in the cloud and pushed to the sensor; exploit mitigation stops memory tricks; EDR records everything like a DVR to catch what prevention misses.

④ How a detection is scored — and how to tune policy without drowning

Every signal — an IOC hit, an ML verdict, an IOA chain, an exploit attempt — is scored for confidence and assigned a severity (informational through critical). Falcon then surfaces a detection in the console with the full process tree, the technique (often mapped to MITRE ATT&CK), the user and the host, so an analyst can triage by severity rather than wade through raw events.

Detection vs prevention is the key toggle

This is the setting people get wrong. Detection means Falcon watches and alerts; prevention means it actively blocks, quarantines or kills. They are separate controls, set per policy. Smart rollout: enable sensor visibility and detection first, watch what fires, then promote trusted detections to prevention. Turn on quarantine so blocked files are isolated (note quarantine is not available on Linux), set exploit mitigation, and apply different policies to different groups — finance, developers and production servers each need their own risk balance.

The failure everyone hits is flipping every slider to Extra Aggressive and full prevention on day one — an instant false-positive storm that blocks legitimate tools. Start in detection, baseline, then enforce.

Figure 5 — From signal to surfaced detection
Any signal is scored, given a severity and raised as a detection with the full story — never a blind, contextless alert.From signal to surfaced detectionSignalIOC / ML / IOA hitScoreconfidence assignedSeverityinfo to criticalSurfacedetection in consoleTriageprocess tree +ATT&CK
Any signal is scored, given a severity and raised as a detection with the full story — never a blind, contextless alert.

Priya, a SOC analyst in Hyderabad, faces this

After flipping the new prevention policy to full block with ML on Extra Aggressive, the helpdesk is flooded — a custom finance macro and a developer's build tool are being quarantined as malicious.

Likely cause

Prevention was turned on at maximum sensitivity with no detection-first baseline, so high-confidence-but-benign behaviours got blocked.

Diagnosis

Open the detections list — the 'malicious' hits are internal, signed tools whose behaviour looks attack-like; they are false positives, not a real intrusion.

Falcon console ▸ Endpoint security ▸ Prevention policies + Detections
Fix

Dial ML prevention back to Moderate, keep detection at Aggressive to stay visible, add tuned exclusions or allowlist entries for the verified internal tools, baseline, then re-raise prevention.

Verify

Re-test: the finance macro and build tool run, the helpdesk queue clears, and the detections list now shows only genuine suspicious behaviour.

Prove it from the detection, not a hunch

Never close a Falcon ticket on 'looks fine'. The detection shows the scored confidence, the severity, the full process tree and the MITRE ATT&CK technique. That single view tells you whether it was an IOC hit, an ML verdict or a behavioural IOA — and whether it was a real attack or a false positive to tune.

Quick check · Q4 of 10 · Analyze

What is the safest way to roll out a new Falcon prevention policy?

Correct: d. Detection and prevention are separate toggles. Going straight to maximum prevention causes a false-positive storm that blocks legitimate tools. Start in detection, watch what fires, then promote trusted detections to active blocking.
👉 So far: Every signal is scored for confidence and severity and surfaced as a detection with the process tree and ATT&CK technique. Detection vs prevention are separate toggles — start in detection, baseline, then enforce.

🤖 Ask the AI Tutor

Tap any question — instant, scoped to this lesson. No login, no waiting.

Pre-curated from vendor docs + community Q&A, scoped to this lesson. For a live prod issue, paste your export into chat.techclick.in.

📝 Wrap-up assessment — six more

You've answered 4 inline. Six left. 70% (7 of 10) marks the lesson complete on your profile. Tap Submit all answers at the end.

Q5 · Remember

An IOC (Indicator of Compromise) is best described as…

Correct: b. An IOC is a leftover artifact — a hash, IP or domain — discovered after the fact. The chain of actions an attack performs is an IOA; ML sensitivity is a separate policy setting.
Q6 · Understand

Why is a behaviour-based IOA harder for an attacker to evade than a signature/IOC?

Correct: c. Repacking a file produces a new hash in seconds, defeating signatures. But an attack must still execute, hide, persist and call out — IOAs watch those required actions, so they catch the attack regardless of the file.
Q7 · Apply

A never-before-seen executable with no known hash tries to run on an offline laptop. What gives Falcon a chance to block it?

Correct: d. On-sensor ML runs on the device and can score and block an unknown file even with no cloud connection. Cloud ML adds depth when online, but local ML is what protects an offline endpoint against unknown malware.
Q8 · Analyze

How do AI-powered IOAs combine cloud and sensor?

Correct: a. Cloud-native ML models trained on the Security Cloud generate IOAs and share them with the Falcon sensor in real time; the sensor correlates those AI-generated behavioural indicators with local events and file data to decide maliciousness.
Q9 · Evaluate

An interviewer asks why EDR continuous recording matters if NGAV already prevents threats. Strongest answer?

Correct: b. No prevention engine is 100 percent. EDR records endpoint activity like a DVR so fileless or living-off-the-land attacks that evade NGAV are still detected and can be investigated step by step — prevention and recording are complementary.
Q10 · Evaluate

What best explains the relationship between Falcon's detection and prevention settings?

Correct: c. Detection alerts and records; prevention actively blocks, quarantines or kills. They are separate controls with their own ML sensitivity levels, so you can detect aggressively while blocking conservatively — start in detection, baseline, then promote to prevention.
Lesson complete — saved to your profile.
Almost! You need 70% (7 of 10) — re-read the path that tripped you up and tap "Try again".

🧠 In your own words

Type one line: why does CrowdStrike say 'behaviour beats signatures', and where do IOAs, ML and EDR each fit? Then compare with the expert version.

Expert version: Because attackers can change files endlessly but cannot avoid the actions an attack requires. Falcon blocks known-bad with IOCs (hashes/IPs/domains), catches unknown malware with on-sensor and cloud machine learning, and stops fileless and novel attacks with behavioural IOAs — extended by AI-powered IOAs generated in the cloud and pushed to the sensor — plus exploit mitigation for memory tricks. An IOA describes attacker behaviour and intent in real time, while an IOC is only a leftover artifact found after the fact. Because prevention is never perfect, EDR continuously records endpoint activity like a DVR to detect and replay anything that slips past. Every signal is scored for confidence and severity and surfaced as a detection with the full process tree and ATT&CK technique, and you tune it through separate detection-vs-prevention toggles — which is exactly why behaviour-first detection beats signature-only antivirus.

🗣 Teach a friend

Best way to lock it in — explain it in one line to a teammate. Tap to generate a paste-ready summary.

📖 Glossary

Indicator of Attack (IOA)
A description of attacker behaviour and intent — the chain of actions an attack must perform (execute, hide, persist, command-and-control), caught in real time regardless of the file used.
Indicator of Compromise (IOC)
A forensic artifact left behind after an intrusion, such as a file hash, IP address or domain. Useful after the fact, but easy for an attacker to change.
NGAV (Falcon Prevent)
Next-generation antivirus that replaces signature-only AV with IOC/hash blocking, machine learning and behavioural IOAs in one lightweight agent.
On-sensor machine learning
ML that runs locally on the endpoint to score an unknown file's maliciousness, so it can block brand-new malware even when offline.
Cloud machine learning
ML models trained on the Security Cloud's telemetry that add a deeper verdict on files and behaviour and improve continuously without a reinstall.
AI-powered IOAs
Behavioural detections generated by cloud-native ML at runtime and pushed to the sensor, which correlates them with local events to assess maliciousness.
Behavioural prevention
Blocking based on the sequence of actions a process takes (the attack chain) rather than the file itself, catching fileless and living-off-the-land attacks.
Exploit mitigation
Protection against memory-based attacks such as buffer overflows and code injection that never drop a file to disk.
EDR (Falcon Insight)
Endpoint detection and response that records endpoint activity like a DVR, so threats that evade prevention are still detected and can be replayed.
Detection vs prevention
Separate policy toggles: detection watches and alerts; prevention actively blocks, quarantines or kills. Each has its own ML sensitivity level.

📚 Sources

  1. CrowdStrike — IOA vs IOC: Understanding the Differences. crowdstrike.com/cybersecurity-101/threat-intelligence/ioa-vs-ioc
  2. CrowdStrike Blog — Introducing AI-Powered Indicators of Attack (cloud ML generating and pushing IOAs to the sensor). crowdstrike.com/blog/introducing-ai-powered-indicators-of-attack-ioas
  3. CrowdStrike — Falcon Prevent: next-generation antivirus (machine learning, exploit blocking and IOA behavioural protection). crowdstrike.com/products/endpoint-security/falcon-prevent
  4. CrowdStrike — What is EDR? Endpoint Detection and Response Defined (DVR-like continuous recording). crowdstrike.com/cybersecurity-101/endpoint-security/endpoint-detection-and-response-edr
  5. CrowdStrike — Falcon Insight EDR data sheet: continuous endpoint visibility and automated detection. crowdstrike.com
  6. CrowdStrike Falcon docs — Prevention policy settings: detection vs prevention, ML sensitivity levels, quarantine and exploit mitigation. falcon.crowdstrike.com

What's next?

Got how detections are made? Next, zoom out to the Falcon architecture — the single lightweight sensor, the cloud-native Security Cloud and the Threat Graph that correlates trillions of events behind every one of these detections.