🧰 AI Log Analysis Tools

Wazuh, Elastic SIEM, Sentinel, Splunk UEBA — what they cost, what they require, what they actually do

🎯 How to Read This Page

This page covers the tools you can actually deploy — from free open-source options a single sysadmin can stand up on a weekend, to enterprise platforms that require a team and a budget conversation with management.

Each tool gets an honest assessment: what AI capabilities it actually has (not what the marketing page says), what it realistically costs to run, what skills and infrastructure it requires, and where it falls short. The goal is to help you match the right tool to your actual environment, not the environment the vendor assumes you have.

No tool in this section is universally right or wrong. A $0 Wazuh deployment tuned by someone who knows their environment will outperform a $200K Splunk deployment that nobody has time to manage. Fit matters more than features.

📊 The Landscape at a Glance

Tool Cost model AI capability Right for
Wazuh Free / open-source ML anomaly detection, rule correlation Small-to-medium shops, budget-conscious, self-hosted
Elastic SIEM Free tier + paid ML ML jobs, anomaly scoring, behavioral analytics Shops already running ELK, developers comfortable with configuration
Microsoft Sentinel Pay-per-GB ingested UEBA, Copilot for Security integration, ML analytics Azure shops, Microsoft-heavy environments
Splunk UEBA Licensed (expensive) Deep behavioral analytics, peer group analysis, kill chain detection Enterprise, dedicated SOC, large log volumes
OpenSearch + custom Free / infrastructure cost Whatever you build Engineers who want full control and have the time to build it

🦷 Wazuh — The SysAdmin's SIEM

🦷 Wazuh

💰 Free & Open Source Self-hosted Agent-based Active community

Wazuh is a security monitoring platform built on top of OSSEC, extended significantly over the years into a full SIEM with its own indexer, dashboard, and manager. It is the most realistic starting point for a sysadmin shop without a dedicated security budget.

✔ Strengths
  • Genuinely free — no license fees, no data caps
  • Agent runs on Linux, Windows, macOS
  • Built-in rules for common attack patterns
  • File integrity monitoring included
  • Active response: can block IPs automatically
  • Solid documentation, active community
  • Integrates with VirusTotal, TheHive, PagerDuty
✘ Limitations
  • ML features require tuning investment
  • Dashboard is functional, not polished
  • High-volume environments need real hardware
  • Alerting configuration has a learning curve
  • Behavioral baselines less sophisticated than Splunk UEBA

What the AI/ML actually does in Wazuh

Wazuh's ML capability centers on its anomaly detection module, which uses statistical analysis to baseline normal behavior per agent and flag deviations. It also does threat intelligence correlation — matching observed indicators against known bad IP/hash/domain feeds.

The rule correlation engine is not traditional ML but is worth understanding: it chains multiple lower-severity events into higher-severity alerts. Five failed logins followed by a success from the same IP is not three separate events — it is one brute-force-then-success alert.

Getting started on a single server

# Wazuh all-in-one install (manager + indexer + dashboard)
# Minimum: 4 cores, 8GB RAM, 50GB disk for a small deployment

curl -sO https://packages.wazuh.com/4.7/wazuh-install.sh
curl -sO https://packages.wazuh.com/4.7/config.yml

# Edit config.yml — set your node name and IP
# Then run the installer
sudo bash wazuh-install.sh -a

# Dashboard available at https://your-server-ip
# Default credentials printed at end of install
# Deploy agent on a Linux host you want to monitor
# Run on the monitored host, pointing at your Wazuh manager

curl -sO https://packages.wazuh.com/4.7/wazuh-agent.sh
sudo WAZUH_MANAGER='your.manager.ip' bash wazuh-agent.sh
sudo systemctl enable --now wazuh-agent
Start small. Deploy Wazuh on your manager plus two or three agents before scaling out. Get familiar with the alert flow, tune the noise down, then expand. A Wazuh deployment that generates 500 alerts a day that nobody reads is worse than no Wazuh at all.

🔍 Elastic SIEM — Power With a Price of Complexity

🔍 Elastic SIEM (Elastic Security)

💰 Free tier available ML requires Platinum/Enterprise license Self-hosted or Elastic Cloud

If you are already running an ELK stack for log aggregation, Elastic SIEM is the natural path to AI-assisted detection. It sits on top of Elasticsearch and adds security-specific detection rules, ML anomaly jobs, and UEBA (User and Entity Behavior Analytics) at the paid tiers.

✔ Strengths
  • Reuses existing ELK infrastructure
  • Excellent ML anomaly detection at paid tiers
  • Detection rules community (Elastic and open)
  • Strong visualization and alerting
  • Timeline investigation interface is genuinely good
  • Scales to very large log volumes
✘ Limitations
  • ML features locked behind paid license
  • Complex to tune and maintain at scale
  • Resource-hungry — needs real infrastructure
  • Steep learning curve for advanced features
  • Licensing costs can escalate with data volume

ML jobs that matter for intrusion detection

Elastic's ML module runs background jobs that continuously analyze your data streams. The most useful for a SysAdmin environment:

ML Job What it detects
auth_rare_source_ip_for_a_user Login from an IP this user has never used — covers the lateral movement and credential theft scenarios from Page 2
auth_rare_hour_for_a_user Login at an unusual hour for this specific user
suspicious_login_activity Combination of factors: new user agent, new source, unusual time
high_count_network_denies_source Host generating unusual firewall deny volume — scanning behavior
rare_process_by_host Process execution that has never been seen on this host before

Enabling ML jobs

# Via Kibana Dev Tools or API — enable a pre-built ML job
# First ensure data feeds are configured (Filebeat or agent sending to ES)

# PUT request to open the job
curl -X POST "localhost:9200/_ml/anomaly_detectors/auth_rare_source_ip_for_a_user/_open" \
  -H "Content-Type: application/json"

# Start the datafeed
curl -X POST "localhost:9200/_ml/datafeeds/datafeed-auth_rare_source_ip_for_a_user/_start"

# Check job status
curl -s "localhost:9200/_ml/anomaly_detectors/auth_rare_source_ip_for_a_user" \
  | python3 -m json.tool | grep "state\|processed"

☁️ Microsoft Sentinel — If You Live in Azure

☁️ Microsoft Sentinel

💰 Pay-per-GB ingested Azure-hosted No infrastructure to manage

Sentinel is Microsoft's cloud-native SIEM. If your environment is Azure-heavy or Microsoft 365-heavy, Sentinel has native connectors that make ingestion nearly effortless. The AI capabilities are genuine and improving rapidly, especially with Copilot for Security integration.

✔ Strengths
  • Zero infrastructure to manage
  • Native Azure AD, M365, Defender connectors
  • UEBA built in at no extra license cost
  • Copilot for Security integration (natural language queries)
  • Good KQL query language once you learn it
  • Scales automatically
✘ Limitations
  • Cost scales with data volume — can surprise you
  • Linux/on-prem ingestion requires agent configuration
  • Vendor lock-in to Azure ecosystem
  • KQL has a learning curve
  • Less useful if you are not Azure-centric

Cost reality check

Sentinel charges per GB of data ingested. Microsoft 365 and Azure-native logs ingest free. Everything else costs money. Before you connect your Linux syslog feeds, do the math:

# Estimate your daily log volume before committing
# Run this on your log sources for a week to get a realistic number

journalctl --since "24 hours ago" | wc -c
# Bytes per day from systemd journal

du -sh /var/log/nginx/access.log
# Size of today's web log

# Multiply daily GB by ~$2.46/GB (Sentinel pricing as of 2026)
# 10GB/day = ~$25/day = ~$750/month just for ingestion
# Use commitment tiers if you know your volume — significant discount

UEBA in Sentinel

Sentinel's UEBA module builds behavioral profiles automatically once enabled. It correlates identity data from Azure AD with activity across connected sources and produces investigation priority scores per user and entity. The Copilot for Security integration lets you query it in plain English:

Copilot for Security sits on top of Sentinel (and other Microsoft security products) and lets analysts ask questions like "show me all anomalous activity for this user in the last 7 days" or "summarize the attack timeline for this incident" in natural language. It is genuinely useful for incident investigation, less so for routine monitoring. It carries its own per-seat cost.

📈 Splunk UEBA — Enterprise Grade, Enterprise Cost

📈 Splunk UEBA

💰 Licensed — significant cost Self-hosted or Splunk Cloud Requires Splunk Enterprise

Splunk UEBA is the most sophisticated behavioral analytics platform in this list. It uses unsupervised machine learning to build behavioral models across users, devices, and applications, with kill chain detection that correlates activity across multiple stages of an attack. It is also the most expensive option by a significant margin and requires a dedicated Splunk deployment to run on top of.

✔ Strengths
  • Best-in-class behavioral analytics depth
  • Peer group analysis across user populations
  • Kill chain correlation across multiple stages
  • Threat timeline visualization is excellent
  • Extensive integration ecosystem
  • Strong support and professional services available
✘ Limitations
  • Expensive — licensing conversations involve procurement
  • Requires Splunk Enterprise as the base platform
  • Needs dedicated staff to operate well
  • Overkill for environments without a SOC
  • Implementation projects are measured in months

If your shop does not have a dedicated security operations function, Splunk UEBA is probably not the right fit. It is built for environments where someone is watching the dashboard full-time. The sophistication of the detection is wasted if nobody is actioning the alerts.

Free Splunk option: Splunk Free tier allows 500MB/day of indexing. Not useful for production security monitoring, but enough to evaluate whether Splunk's interface and query language (SPL) work for your team before committing to licensing.

🔨 Open-Source Alternatives Worth Knowing

Beyond the major platforms, several open-source tools deserve mention for shops that want to build their own stack:

OSSEC

The ancestor of Wazuh. Still maintained, lighter weight, useful for environments where Wazuh's full stack is too heavy. Host-based intrusion detection, file integrity monitoring, log analysis. No ML, but solid rule-based detection.

Zeek (formerly Bro)

Network traffic analysis framework. Does not analyze host logs — it analyzes packets. Produces structured logs of network activity that can be fed into any SIEM or analyzed directly. Excellent for detecting lateral movement at the network layer, C2 beaconing, and data exfiltration patterns that host logs miss entirely.

# Zeek produces structured logs per protocol
# conn.log — all network connections with bytes transferred
# dns.log  — all DNS queries (exfiltration often uses DNS)
# http.log — all HTTP with user agents, URIs, response codes
# ssh.log  — SSH sessions with direction and bytes

# Feed Zeek logs to sgpt for behavioral analysis
tail -200 /var/log/zeek/current/conn.log | \
  sgpt "These are network connection logs. Identify any unusual outbound
  connections, high data transfer volumes, or connections to unexpected
  destinations. Flag anything worth investigating."

Suricata

Network IDS/IPS with rule-based detection plus some behavioral capabilities. Complements host-based tools by watching the wire. Good at detecting known attack tools and C2 frameworks at the network layer. Runs inline (IPS mode) or passively (IDS mode) depending on your preference.

Graylog

Open-source log management with alerting and some anomaly detection. Not as security-focused as Wazuh or Elastic SIEM, but a solid log aggregation platform that pairs well with AI analysis pipelines. The free tier is genuinely functional for small environments.

🎯 Choosing the Right Tool for Your Environment

Your situation Recommended starting point Why
Solo sysadmin, limited budget, Linux-only environment Wazuh Free, full-featured, manageable by one person who invests in learning it
Already running ELK stack Elastic SIEM (free tier first) Reuses existing infrastructure and team knowledge
Azure-heavy, Microsoft 365 shop Microsoft Sentinel Native connectors make ingestion easy; UEBA built in
Enterprise with dedicated SOC Splunk UEBA Depth of behavioral analytics justifies cost at scale with staff to use it
Want network-layer visibility alongside host logs Add Zeek to any of the above Host and network logs together catch what either misses alone
Not ready for any of the above yet curl/sgpt pipeline from Page 1 Start getting value from AI log analysis today with zero new infrastructure
A tool you deploy and ignore is worse than no tool. Every SIEM generates alerts. Alerts that are never actioned train your team to ignore them. Before deploying anything here, have a plan for who reviews alerts, how often, and what the escalation path is. A well-tuned Wazuh instance with a weekly review beats an enterprise SIEM with alert fatigue. Page 4 covers this problem in depth.

➡️ Where to Go Next

You have the tool landscape. Page 4 is the one that puts everything in perspective — what all of these tools miss, why the detection asymmetry favors attackers, and the honest accounting of AI security's limits.