What to look for in a next-generation antivirus (NGAV)
Back to blog

What to look for in a next-generation antivirus (NGAV)

6/7/2026 · 6 min · Cybersecurity

Next-generation antivirus (NGAV) was designed to close security gaps left by legacy solutions. Traditional signature-based antivirus struggles against modern threats such as zero-day malware, fileless attacks, and advanced ransomware.

This guide goes beyond conceptual overviews - it provides tools, metrics, practical implementation steps, and objective criteria to select and validate an NGAV suited to your environment.


Why NGAV matters: real breach data#

Before evaluating tools, understand what happens when endpoint defense is insufficient:

CaseYearImpactRelevant Root Cause
Maersk (NotPetya)2017$300M in losses, 45,000 PCs reinstalledTraditional AV failed to detect fileless malware + lateral propagation
Capital One2019100M records exposed, $80M in finesMisconfigured WAF + no behavioral detection
SolarWinds202018,000 organizations compromisedNGAV failed to detect supply chain backdoor for 9 months
Kaseya VSA20211,500+ companies via MSPEPP/NGAV failed to block REvil ransomware via trusted process

In every case, an NGAV with behavioral detection, parent/child process analysis, and SIEM integration would have detected anomalies before critical damage occurred.


NGAV vs traditional antivirus: technical comparison#

CharacteristicTraditional AVNGAV
Detection mechanismStatic signaturesML + Behavioral + Signatures + IOAs
Zero-day threats❌ Limited (no signature exists)✅ Detection via behavioral patterns
Fileless malware❌ Not detected (no file on disk)✅ Monitors in-memory and PowerShell execution
Ransomware⚠️ Partial (post-execution detection)✅ Blocks before encryption via IOA + behavior
UpdatesManual/periodic (hours to days)Continuous via cloud (minutes)
PerformanceLight (signatures only)Moderate (ML + telemetry)
False positivesLow (signature = certainty)Medium initially (ML requires tuning)
Offline detection✅ Local signatures✅ Local ML model (offline mode)
Cost$$$ to $$$
EDR/SIEM integration❌ Limited or none✅ Native via API

NGAV tools in the market#

Enterprise#

SolutionDetection MechanismKey Differentiator
CrowdStrike FalconML + IOA + cloud Threat GraphReal-time global threat intelligence; autonomous response
SentinelOne SingularitySignature-less behavioral AIAutomatic ransomware rollback; autonomous response
Microsoft Defender for EndpointML + Azure AD integration + IOCNative M365 stack; Conditional Access integration
Palo Alto Cortex XDRML + unified XDR correlationEPP + EDR + XDR in a single platform

SMB / mid-market#

SolutionKey Differentiator
Bitdefender GravityZoneNGAV + centralized console; strong cost/benefit ratio
Sophos Intercept XSignature-less deep learning; CryptoGuard anti-ransomware
Kaspersky Endpoint SecurityNGAV + integrated disk encryption

Open source / hybrid#

SolutionNGAV ComponentLimitation
WazuhBehavioral detection + FIM + SIEMRequires own infrastructure; no advanced native ML
OSSECHIDS + file integrity monitoringNo modern UI; detection limited to log analysis
ClamAVSignatures onlyNot NGAV - covers known threats only

NGAV evaluation metrics#

Before signing a contract, require this data from the vendor and validate it during your POC:

Detection#

MetricMinimum TargetHow to Measure
Known malware detection rate> 99%AV-TEST / AV-Comparatives benchmarks
Zero-day detection rate> 95%MITRE ATT&CK Evaluations simulation
Time to detect new threat< 1 hourDetection timestamp vs event creation timestamp
False positive rate< 1%Monitor alerts for 2 weeks in audit mode

Performance#

MetricMinimum Target
CPU at idle< 5%
CPU during full scan< 30%
Agent RAM usage< 500 MB
Full scan completion time< 30 minutes
Agent availability> 99.9%

Operations#

MetricMinimum Target
Signature/model update time< 5 minutes
MTTD (Mean Time to Detection)< 1 hour
MTTR (Mean Time to Response)< 4 hours
Active agent coverage> 95% of fleet

Technical implementation: Wazuh NGAV + XDR (open source)#

For environments requiring an auditable, license-free solution, Wazuh provides behavioral detection, FIM, and SIEM integration.

Agent installation (ubuntu/debian)#

# 1. Download and install Wazuh agent 4.7.x
curl -sO https://packages.wazuh.com/4.x/apt/pool/main/w/wazuh-agent/wazuh-agent_4.7.3-1_amd64.deb
sudo dpkg -i wazuh-agent_4.7.3-1_amd64.deb

# 2. Configure the Manager address
sudo sed -i 's/<address>MANAGER_IP<\/address>/<address>192.168.1.100<\/address>/' \
  /var/ossec/etc/ossec.conf

# 3. Enable process monitoring and FIM in ossec.conf
# Add inside <syscheck>:
#   <frequency>3600</frequency>
#   <directories check_all="yes">/usr/bin,/usr/sbin,/bin,/sbin</directories>

# 4. Enable and start the agent
sudo systemctl enable wazuh-agent
sudo systemctl start wazuh-agent

# 5. Verify status
sudo systemctl status wazuh-agent
sudo /var/ossec/bin/wazuh-control status

Installation on Rocky Linux / CentOS#

sudo rpm --import https://packages.wazuh.com/key/GPG-KEY-WAZUH
sudo bash -c 'cat > /etc/yum.repos.d/wazuh.repo << EOF
[wazuh]
gpgcheck=1
gpgkey=https://packages.wazuh.com/key/GPG-KEY-WAZUH
enabled=1
name=EL-\$releasever - Wazuh
baseurl=https://packages.wazuh.com/4.x/yum/
protect=1
EOF'
sudo yum install wazuh-agent -y
sudo sed -i 's/MANAGER_IP/192.168.1.100/' /var/ossec/etc/ossec.conf
sudo systemctl start wazuh-agent && sudo systemctl enable wazuh-agent

Detection validation with EICAR test file#

# Test detection with the EICAR standard string (harmless, detected by all AVs)
echo 'X5O!P%@AP[4\PZX54(P^)7CC)7}$EICAR-STANDARD-ANTIVIRUS-TEST-FILE!$H+H*' \
  > /tmp/eicar-test.txt

# Verify Wazuh generated alerts (wait ~30s)
sudo tail -f /var/ossec/logs/alerts/alerts.log | grep -i eicar

# Clean up after testing
rm /tmp/eicar-test.txt

Suspicious behavior simulation (IOA detection)#

# Simulate suspicious child process execution (common malware behavior)
# ONLY in an isolated test environment
bash -c "whoami; id; cat /etc/passwd | head -5"

# Check generated Wazuh alert
sudo tail -f /var/ossec/logs/alerts/alerts.log | grep -E "level|description" | head -20

Selection criteria: full checklist#

1. Functionality#

- [ ] ML/AI-based detection (not signatures alone)
- [ ] Real-time behavioral analysis (IOAs)
- [ ] Zero-day and fileless malware protection
- [ ] Ransomware blocking before encryption
- [ ] In-memory attack detection
- [ ] Native EDR/XDR integration

2. Agent security#

- [ ] Agent-to-console communication encrypted (TLS 1.2+)
- [ ] MFA mandatory for management console
- [ ] Immutable audit logs (WORM or signed)
- [ ] Certifications: ISO 27001, SOC 2 Type II, Common Criteria
- [ ] Tamper protection (agent cannot be disabled by a local process)

3. Performance and compatibility#

- [ ] CPU impact < 5% at idle
- [ ] Linux (LTS kernel), Windows Server, macOS support
- [ ] Container support (Docker, Podman, Kubernetes)
- [ ] Offline protection with local ML model
- [ ] Automatic rollback on agent failure

4. Operations#

- [ ] Centralized management console with RBAC
- [ ] Granular policies by group/OU
- [ ] Automated reports with configurable frequency
- [ ] Real-time alerts with severity and context
- [ ] Documented REST API for automation

5. Integration#

- [ ] Event forwarding to SIEM via syslog/CEF/LEEF
- [ ] SOAR integration for automated playbooks
- [ ] MITRE ATT&CK framework compatibility
- [ ] External threat intelligence feeds (STIX/TAXII)

6. Total cost of ownership (TCO)#

- [ ] Per-endpoint/year cost (license)
- [ ] Infrastructure cost (SaaS console vs on-premise)
- [ ] Deployment and training costs
- [ ] Technical support cost (24/7 vs business hours)
- [ ] Price scalability for fleet growth

NGAV agent performance impact#

ResourceTraditional AVNGAVNotes
CPU (idle)1–3%2–5%NGAV has ML model overhead
CPU (full scan)20–50%15–40%NGAV often more efficient due to cloud offloading
CPU (behavioral detection)-5–15% additionalOnly during suspicious process analysis
RAM (agent)100–300 MB150–400 MBLocal ML models consume more
RAM (ML models)-200–500 MBLoaded on demand
Disk (logs)100–500 MB/day200–800 MB/dayBehavioral telemetry is more verbose
Disk (ML cache)-1–2 GBLocal models for offline detection
Network (telemetry)1–5 MB/day5–20 MB/dayNGAV sends more context to cloud
Network (updates)10–50 MB/week20–100 MB/weekML models are larger than signatures
# Monitor NGAV agent consumption in real time
top -p $(pgrep wazuh-agentd) -b -n 3 | grep -A2 wazuh

# Disk usage of logs:
du -sh /var/ossec/logs/

# I/O during scan:
iotop -p $(pgrep wazuh-agentd) -b -n 5

Regulatory compliance#

FrameworkControlPractical NGAV Obligation
PCI-DSS v4.0Control 5 (anti-malware)NGAV/EDR on all cardholder data systems; annual control verification; 12-month log retention
HIPAA§164.312(a)(1)Anti-malware on PHI systems; access auditing; fines up to $1.9M per incident
LGPD (Brazil)Art. 46Technical measures against unauthorized access; 72h ANPD notification; fines up to 2% revenue (max R$50M)
ISO 27001:2022A.8.7 + A.12.6.1Documented malware management policy; vulnerability management process; periodic alert review
SOC 2 Type IICC6.8 (Logical Access)Documented malware controls; evidence of continuous monitoring; annual effectiveness report

Core NGAV capabilities#

When evaluating vendors, look for:

  1. AI and Machine Learning (ML)

The ability to process large data sets, detect complex attack patterns, and identify malicious behavior in real time, ideally using Indicators of Attack (IOAs) and exploit mitigation.

  1. Unknown threat protection

Defense against known and unknown threats (especially zero-day attacks) through behavioral analytics instead of relying only on static signatures.

  1. Policy customization

Support for allowlists and blocklists so security teams can adapt controls to business applications and internal risk policies.

  1. Cloud-native architecture

Faster deployment, continuous updates, and lower endpoint impact, while keeping protection consistent even when devices are offline.


Unified XDR: multi-layer correlation#

The market direction is clear: isolated NGAV is not enough. XDR (Extended Detection and Response) unifies endpoint, network, cloud, and identity into a single kill chain view.

Complete Kill Chain with XDR:
Email phishing → Endpoint (NGAV) → Network (NDR) → Cloud (CASB) → Identity (IdP)
      ↕                ↕                 ↕                ↕               ↕
            XDR correlates all signals into a single, unified alert

Solutions exemplifying this convergence: CrowdStrike Falcon XDR, SentinelOne Singularity XDR, Microsoft Defender XDR.

Generative AI in security operations#

Emerging use cases:
- Automatic incident analysis: "What happened, when, and what is the impact?"
- Natural language report generation for C-level stakeholders
- Contextualized remediation suggestions (which patch, which config, which rule)
- Alert triage with automatic priority classification

Market examples:
- CrowdStrike Charlotte AI
- SentinelOne Purple AI
- Microsoft Copilot for Security

Zero trust + NGAV: dynamic defense in depth#

Zero Trust integration with NGAV:
1. Non-compliant endpoint (NGAV detects anomaly)
   → Signal sent to Identity Provider
   → Session automatically revoked
   → Access blocked until remediation

Typical implementation:
- CrowdStrike Falcon ↔ Okta/Azure AD (Conditional Access)
- SentinelOne ↔ Zscaler (network segmentation)
- Wazuh ↔ OpenID Connect + Policy Engine

NGAV for containers and Kubernetes#

# Protect containerized workloads with Wazuh
# The Wazuh agent on the host automatically monitors container events

# Check container events in the alerts log
sudo grep -i "docker\|container\|namespace" /var/ossec/logs/alerts/alerts.json \
  | python3 -m json.tool | head -30

# Configure Docker event monitoring in ossec.conf:
# <localfile>
#   <log_format>syslog</log_format>
#   <location>/var/lib/docker/containers/*/*-json.log</location>
# </localfile>

What to expect from a top-tier solution#

A strong NGAV should be:

In practice, the better the integration with your security ecosystem, the higher your overall prevention and response maturity. Selecting the right NGAV starts with the objective criteria presented in this guide - and ends with a rigorous POC in your own environment.

Was this article helpful?

Leave a quick reaction to help prioritize future technical guides:

CC BY-NC

This post is licensed under CC BY-NC.

Comments

Join the discussion below.

0 comments