Fortinet FortiAnalyzer: Detection Rules and Correlation

In FortiAnalyzer security-operations workflows, detection logic is built around event handlers that transform raw logs into events analysts can investigate. The important design choice is not how many rules can be created; it is how precisely the rules represent behavior that deserves operational attention. FortiAnalyzer supports basic event handlers, where any matching rule can generate an event, and correlation event handlers, where multiple rules must satisfy a defined sequence or logical relationship. Used carefully, that model can reduce manual search. Used carelessly, it can create a second layer of alert noise on top of already noisy telemetry.

Fortinet security operations should treat detection rules as maintained analytical code. They have scope, input assumptions, grouping keys, time relationships, severity, notification behavior, and downstream consequences. Every one of those elements can become wrong as networks, products, identities, and attacker behavior change. The strongest rule program therefore resembles mature cloud-native SIEM architecture practice: normalize what the rule assumes, limit the population it evaluates, correlate only fields with defensible meaning, test against known benign and malicious sequences, and measure whether the resulting event improves an analyst’s decision.

Choose basic or correlation logic based on the behavior

A basic event handler is appropriate when one log condition is already strong enough to justify an event. FortiAnalyzer evaluates the rules in a basic handler with OR semantics, so a match on any one can generate the event. That makes basic handlers useful for high-confidence conditions such as a specific critical administrative action, a device health condition, or a known signature that requires immediate review.

Correlation handlers solve a different problem. They are useful when meaning emerges from sequence or combination: an authentication pattern followed by a privileged action, reconnaissance followed by exploitation, or several events sharing an entity within a time window. FortiAnalyzer supports operators such as AND, AND_NOT, OR, FOLLOWED_BY, and NOT_FOLLOWED_BY, allowing a rule to encode relationships rather than isolated matches.

The decision should be driven by what makes the behavior suspicious. If one record is enough, correlation adds complexity without value. If several weak signals become strong only when linked, a basic handler usually creates noise.

Predefined handlers are useful starting points, especially in Security Fabric ADOMs where FortiAnalyzer exposes detections for FortiGate, FortiSandbox, FortiMail, and FortiWeb. They should still be reviewed before enablement. A predefined rule may assume log fields, product features, or traffic patterns that do not match the local environment. Clone when the detection concept is useful but the operational scope needs to change.

Document why a custom handler exists separately from the vendor default. That record helps future engineers decide whether a product update makes the customization unnecessary or whether local risk still requires different logic.

Scope inputs before writing the condition

Event handlers operate on data selected from devices, subnets, filters, and other criteria. That scope is part of the detection, not an administrative afterthought. A rule that works well for internet-facing firewalls may become meaningless when applied to internal lab devices or test traffic. Likewise, a correlation that assumes stable user identity may fail on shared service accounts.

Create data selectors around the environment in which the detection is valid. Scope by device class, network segment, role, application, or other stable boundary before using increasingly complicated conditions to suppress irrelevant records. Narrow scope is often more understandable than a long exception list.

Network security fundamentals matter to detection engineering because rules inherit the architecture around them. If asset purpose, administrative boundaries, and identity ownership are unclear, the rule designer ends up encoding guesses.

Correlation keys determine whether events belong together

Correlation is only useful when the fields used to join rules represent the same real-world entity. Source IP may identify a workstation in one environment and a NAT gateway in another. Username may be meaningful for interactive activity but misleading for service-to-service actions. Destination address may change behind load balancers. The correlation criteria should be chosen with those semantics in mind.

FortiAnalyzer exposes group-by fields and allows correlation criteria to connect fields between rules. The design task is to decide which relationship is trustworthy enough to create one event. Analysts should be able to explain why the selected key proves that two records belong to the same activity chain.

Test correlation keys against shared infrastructure, roaming users, proxies, NAT, DHCP churn, and other conditions that collapse or change identity. A detection that seems precise in a sample dataset can produce misleading chains when deployed across the full environment.

Correlation also depends on event normalization. If two devices represent usernames, zones, applications, or addresses differently, the rule may fail to join activity that is actually related. Before adding more logic, inspect the raw values that populate the chosen fields across every data source in scope.

When NAT or proxy infrastructure obscures origin, consider whether another field—device identifier, user, session, hostname, or a combination—provides a stronger relationship. A correlation that is mathematically valid but semantically weak can mislead analysts more than no correlation at all.

Time windows should reflect attacker and system behavior

Sequence operators such as FOLLOWED_BY become meaningful only with a time limit. That limit should reflect how quickly the behavior can occur and how long normal business processes may take. Too short, and the rule misses distributed or delayed activity. Too long, and unrelated events accumulate into false correlations.

Build the window from evidence. Review historical incidents, baseline administrative workflows, automation timing, token lifetimes, batch schedules, and normal user behavior. If the detection targets password spraying followed by account use, the useful window may differ substantially from a rule that targets malware execution followed by outbound communication.

Time also affects operational volume. Broad windows hold more candidate state and can cause a large environment to correlate activity that only appears related. Rule tuning should therefore consider computational and analytical cost together.

Noise reduction must not erase the attack

Suppression and exceptions are necessary, but each one is a policy decision about what evidence the SOC will ignore. Excluding a scanner, service account, management subnet, or maintenance window may reduce false positives while also hiding attacker behavior that deliberately uses the same infrastructure.

SIEM triage should reconstruct why an alert was unhelpful before tuning changes are made. Change the narrowest condition that explains the problem, and avoid broad “known good” exclusions unless ownership and behavior are truly controlled.

Track exceptions with an owner and expiration or review date. Temporary changes tend to become permanent because they solve an immediate operational pain. A quarterly rule review should identify exceptions that no longer reflect the environment.

Noise reviews should sample both alerts and non-alerting events. Looking only at generated alerts encourages tuning toward the current output while hiding missed behavior. Periodically replay or search for known attack-like sequences and confirm the handler still fires after exclusions are added.

Measure alert concentration by host, user, subnet, and device. A rule that produces most of its volume from one maintenance system may need a scoped exception; a rule that suddenly shifts to a new population may be signaling an environmental change worth investigating.

Attach context and response deliberately

FortiAnalyzer event handlers can reference MITRE ATT&CK technique IDs, notification profiles, and automation mechanisms such as Automation Stitch integration. Those features are valuable when they describe what the rule actually detects and what response is safe.

MITRE mapping should not be decorative. If an event claims a technique, analysts should understand which observable behavior supports that mapping. Additional threat-intelligence enrichment can raise or lower confidence when it changes the decision, but attaching reputation data to every event does not automatically improve detection quality.

Automation should start with low-risk actions such as enrichment, ticket creation, tagging, or evidence capture. Blocking or isolation requires stronger confidence and rollback design. A detection rule that is acceptable for human review may not be safe enough to trigger prevention.

Operate rules as a versioned detection program

Every production rule should have an objective, owner, data dependencies, expected event volume, known false-positive modes, test cases, and a review date. When the environment changes, compare the new event distribution to the old one. Sudden silence can be as important as sudden noise.

FortiAnalyzer sits inside the wider Fortinet ecosystem, so product upgrades, log-schema changes, new FortiGate policies, FortiMail or FortiSandbox integrations, and ADOM changes can all affect rule behavior. Detection owners need change visibility before those changes become unexplained alert drift.

A mature program measures whether rules lead to useful decisions. Track precision, investigation time, repeated false-positive causes, events that never receive action, and detections that were missing from real incidents. Those measures align the rule set with security-operations resilience rather than with a vanity count of enabled detections.

Version rule changes just as teams version firewall policy or infrastructure code. Record the condition before and after, the test dataset, expected volume change, and rollback plan. If an edit unexpectedly suppresses a high-value detection, the SOC should be able to restore the last known-good version quickly.

Rule retirement is also part of quality. Remove or disable detections that no longer have valid data, duplicate stronger analytics, or cover obsolete technology. An ever-growing rule count without retirement makes ownership and review progressively weaker.

Leave a Reply

How It Works

img
Step 1. Choose Exam
on ExamLabs
Download IT Exams Questions & Answers
img
Step 2. Open Exam with
Avanset Exam Simulator
Press here to download VCE Exam Simulator that simulates real exam environment
img
Step 3. Study
& Pass
IT Exams Anywhere, Anytime!