UTMStack

Articles12 min read

What Logs Should Be Sent to a SIEM? 16 Sources in Priority Order

By UTMStack Team

UTMStack blog: SIEM log sources

Every SIEM (security information and event management) project meets the same question in its first week: which logs do we send? “Everything” sounds safe, but it buries the useful events and, with some products, multiplies the bill. “Whatever is easy” is worse. Teams connect the firewall because it speaks syslog, skip the domain controllers because they need an agent, and end up with a great deal of data about traffic and none about who logged in.

There is a better starting point than opinion. In August 2024 the cyber security agencies of nine countries published a joint guide that ranks log sources in order. This article walks through that list, says what to record inside each source, and shows how to roll it out in stages.

The short answer: send logs from the systems an attacker must touch, in this order: your most critical systems and data, internet-facing services and remote access, identity servers such as domain controllers, other critical servers, firewalls and boundary routers, and administrator workstations. Then add privileged tools, data stores, security software, user computers, web proxies, DNS, email and DHCP. In every source, record logins, privilege and account changes, programs started with their full command line, configuration changes and any attempt to clear or stop logging. Keep logs for at least 90 days, and longer where you can.

A rollout order for SIEM log sources

The 16 priorities of the joint government guidance, grouped into four stages. The grouping into stages is ours.

  1. Stage 1What an attacker must touch
    Critical systems and data1Internet-facing services and remote access2Identity servers3Other critical servers4Firewalls and boundary routers5
  2. Stage 2Where the keys are kept
    Administrator workstations6Highly privileged systems7Data repositories8Security and critical software9
  3. Stage 3Where most attacks begin
    User computers10User application logs11Web proxies12DNS13Email servers14
  4. Stage 4Fill the gaps
    DHCP servers15Legacy systems16Hypervisors, printers, application gatewayslower
Numbers are the order in "Best practices for event logging and threat detection" (August 2024). The guide names the last three as lower priority.

Where the list comes from

“Best practices for event logging and threat detection” was written by the Australian Signals Directorate’s Australian Cyber Security Centre. It was released with CISA, the FBI and the NSA in the United States and with agencies in the United Kingdom, Canada, New Zealand, Japan, South Korea, Singapore and the Netherlands. CISA’s page dates it August 21, 2024.

It is useful for two reasons. It is not written by a vendor, so it has no reason to tell you to send more data. And it ranks sources, which most standards avoid. NIST’s log management guide, SP 800-92, dates from September 2006, and its replacement has been a draft since October 2023. The CIS Controls tell you to collect audit logs (safeguard 8.2) and name a few types, but they do not put your systems in order.

The guide gives separate lists for enterprise networks, cloud, operational technology and mobile devices. The enterprise list is the one most teams need first.

The 16 enterprise log sources, in order

#Log sourceTypical examplesWhy it ranks here
1Critical systems and data holdings likely to be targetedCore business application, customer database, file server with sensitive dataIt is what the attacker came for
2Internet-facing services, including remote access, network metadata and their server operating systemVPN gateway, web servers, mail gateway, remote desktop gatewayThe front door, scanned every day
3Identity and domain management serversActive Directory domain controllers, cloud identity provider, single sign-onEvery stolen password is used here
4Any other critical serversBackup servers, database servers, virtualization managementLosing one stops the business
5Edge devices such as boundary routers and firewallsPerimeter firewall, branch routersShows what entered and what left
6Administrative workstationsThe computers your administrators work fromOne of these holds the keys to all the rest
7Highly privileged systemsConfiguration management, monitoring, build and deployment pipelines, vulnerability scanners, password vaultsThey can change every other system
8Data repositoriesDatabases, document stores, code repositoriesWhere copying data in bulk shows up
9Security-related and critical softwareAntivirus and endpoint protection consolesThey already detect; the SIEM connects their alerts to everything else
10User computersLaptops and desktopsWhere phishing and malware land first
11User application logsOffice and collaboration toolsWho opened, shared or downloaded what
12Web proxies used by users and service accountsProxy or secure web gatewayMalware downloads and command channels
13DNS services used by usersInternal DNS resolversAlmost every attack looks up a name
14Email serversMail servers or hosted emailPhishing, forwarding rules, mailbox access
15DHCP serversAddress assignment logsTells you which machine held an address at a given time
16Legacy IT assetsOld systems that cannot be patchedWeak, so watch them more, not less

The guide lists hypervisor hosts, devices such as printers and network components such as application gateways as lower priority.

Two things about the order surprise people. Firewalls are fifth, not first. And user computers are tenth, even though most attacks begin there. The logic is that you start where a compromise does the most harm and where normal activity is easiest to tell from an attack. A domain controller has a small number of administrators and a clear pattern. A thousand laptops do not.

That order is a starting point, not a rule. If your biggest risk is ransomware arriving by email, move user computers and email up. If you run a public web application, your web servers and their databases are items 1 and 2 at once.

What to record inside each source

Connecting a source is half the job. Many systems log too little by default. Whatever the source, make sure it records these events:

Event typeWhy it mattersWhere it is often switched off
Logins, successful and failedBrute force, stolen passwords, logins at odd hoursUsually on, but successes are sometimes dropped to save space
Account and privilege changesNew administrators, new accounts, group changesWindows needs audit policy settings
Programs started, with the full command lineThe clearest view of what an intruder didOff by default on Windows and Linux
Configuration and policy changesFirewall rules, audit settings, security tools disabledVaries by device
Access to sensitive dataWho read or copied itNeeds auditing set on the specific folders or tables
Logs cleared or logging stoppedAn attacker hiding, or a system that failedRarely monitored

The command line deserves a special mention. Attackers increasingly use tools that are already on the system, so there is no malware file to find. The joint guide is blunt about it: “Always enable detailed logging that includes process creation events and command-line auditing.” It suggests writing rules for tools such as powershell.exe, cmd.exe, regedit.exe, curl, systemctl and python when they run encoded commands. CIS safeguard 8.8 asks for command-line logs as well, and safeguards 8.6 and 8.7 ask for DNS queries and web addresses requested.

For the exact settings, see our lists of Active Directory use cases with their Windows event IDs and Linux SIEM use cases with the audit rules that record commands.

Cloud log sources

If part of your environment is in the cloud, the guide gives a second list. The first two items are the same as on premises. The rest are specific to cloud:

  1. Critical systems and data holdings likely to be targeted.
  2. Internet-facing services, including remote access, and their server operating systems.
  3. Use of the accounts that access and administer your cloud services.
  4. Administrative configuration changes.
  5. Creation, deletion and change of every security principal, including its permissions.
  6. Successful and failed sign-ins to third-party services that trust your identity provider.
  7. Logs generated by the cloud services themselves: API calls, network events, compliance and billing events.

In practice items 3 to 5 and 7 are your provider’s audit trail, such as AWS CloudTrail, Azure activity and sign-in logs, or Google Cloud audit logs. For the AWS side, our AWS CloudTrail SIEM use cases guide lists the detections worth running first. Turn on network flow records where the data matters; our VPC flow logs guide explains the format. Billing events are on the list because a sudden rise in spending can be the first sign of stolen cloud credentials.

Operational technology

For industrial systems the guide gives three priorities: devices critical to safety and service delivery (except air-gapped systems), internet-facing devices, and devices reachable across a network boundary. Many controllers cannot produce useful logs at all, so teams usually watch the network around them instead.

How long to keep the logs

The guide sets no fixed period. It says retention should follow your own risk assessment, and that default settings “are often insufficient”. It gives the reason in numbers: it can take up to 18 months to discover an incident, and some malware stays on a network for 70 to 200 days before it causes visible harm.

Compare that with the common minimums. CIS safeguard 8.10 asks for at least 90 days. Mandiant’s M-Trends 2026 report puts the global median time an attacker is present before detection at 14 days, up from 11 the year before. A median means half of intrusions take longer.

Why 90 days is a floor, not a target

How long attackers can go unnoticed, against the most common minimum retention period (days)

Time an attacker goes unnoticedMinimum retention
Median time to detection (M-Trends 2026)14
CIS safeguard 8.10 minimum retention90
Malware before visible harm, upper end (joint guide)200
Time to discover some incidents (joint guide)548
18 months is shown as about 548 days. With 90 days of logs, an investigation into either of the last two cases starts after the evidence is gone.

The guide’s practical advice is two tiers: recent logs in fast storage for searching, older logs in cheaper storage that you can still retrieve. Your industry may set its own period on top. Our guide to SIEM compliance requirements lists them by framework, and the SIEM sizing calculator estimates the storage.

Make the logs usable

The same guide spends as many pages on log quality as on log sources. Four points are worth copying:

  • One clock. Use a trusted time source on every system. The guide prefers UTC, written in ISO 8601 format with milliseconds, for example 2024-07-25T20:54:59.649Z. Without it you cannot put events from two systems in order.
  • Structured format. It recommends a structured format such as JSON and automatic normalization, so that “user” means the same thing in a firewall log and a server log.
  • Send logs off the system quickly. An attacker with administrator rights can erase local logs. A copy in a central place survives.
  • Protect the central copy. Encrypt logs in transit (the guide names TLS 1.3), limit who can change or delete them, keep them on a separate network segment and back them up.

What to leave for later

Sending less at first is not a failure. These sources produce a lot of data for each useful event, so add them once the first two stages work:

  • Allowed-traffic records from internal firewalls. Useful in an investigation, heavy every day. Start with denied traffic, administrator logins and configuration changes.
  • Debug and performance logs. They belong in an operations tool.
  • Every user computer, if you already run endpoint protection. Send that product’s alerts first (item 9), then add full workstation logs.
  • Printers and similar devices. The guide itself ranks them lower.

If your SIEM charges by data volume, this trimming decides your bill, which is why teams on those products often log less than they should. See how SIEM pricing models work.

How UTMStack fits

We build UTMStack, so weigh our view accordingly. Here is how the product lines up with the list, including where it does not.

Priority on the listHow UTMStack collects it
Servers, identity servers, administrator workstations, user computers (1 to 4, 6, 10)Agents for Windows and Linux in the free Community edition. The macOS agent is paid. The Linux agent also turns on command logging through the audit service.
Firewalls and edge devices (5)Built-in parsers and rules for Fortinet FortiGate, Palo Alto Networks, Cisco ASA, Firepower and Meraki, Sophos XG, SonicWall, pfSense and MikroTik, plus NetFlow
Data repositories (8)Oracle Database and GitHub integrations
Security software (9)Kaspersky, ESET, Bitdefender, SentinelOne, Sophos Central and CrowdStrike Falcon
Email and user applications (11, 14)Microsoft 365. There is no Google Workspace parser.
CloudAmazon Web Services, Microsoft Azure and Google Cloud
Virtualization and older platforms (lower priority, 16)VMware ESXi, IBM AIX and IBM AS/400
Web proxies, DNS servers, DHCP servers (12, 13, 15)No dedicated parsers. They can arrive through the generic syslog or JSON inputs, and you write the rules.

That is 33 built-in integrations, listed on the integrations page. More than 600 detection rules mapped to MITRE ATT&CK are checked as each log arrives, before storage. Logs can be kept for up to 5 years, with older data in cold storage that stays searchable; see log management.

The limits, plainly. There are no parsers for industrial control protocols, so the operational technology list above is mostly out of reach. There is no file integrity monitoring. And the proxy, DNS and DHCP sources need your own rules.

One thing does change the decision about what to send. UTMStack is priced per device, never by log volume, so turning on command-line logging or keeping firewall traffic records does not raise the price for that device. For a figure, see pricing or ask for a quote.

Frequently asked questions

Should I send all logs to my SIEM? No. Send the sources on the priority list first and make sure each one records logins, privilege changes, command lines and configuration changes. Add high-volume, low-value sources later, when you have a use for them.

What are the most important log sources for a SIEM? Critical servers and data stores, internet-facing services and remote access, identity servers such as domain controllers, firewalls, and administrator workstations. Those are the first six priorities in the 2024 joint government guidance.

Do I need to send firewall logs to a SIEM? Yes, but they are fifth on the list, not first. Start with denied connections, VPN logins, administrator logins and rule changes. Add records of allowed traffic when you have the capacity.

Do I still need endpoint logs if I have endpoint protection? Send the endpoint product’s alerts to the SIEM first. Add operating system logs from servers and administrator workstations next, because the SIEM can connect them with identity, firewall and cloud events that the endpoint product does not see.

How long should a SIEM keep logs? At least 90 days, the minimum in CIS safeguard 8.10. Longer is better: the joint guidance notes that some incidents take up to 18 months to discover. Check the rules for your industry as well.

How many log sources does a SIEM need? There is no right number. A small company may cover the first six priorities with a few dozen devices. Count coverage of the list, not the number of sources.

Sources