- Learn more
- Threat Intelligence 101: What It Is and How It Works
- Product track
- Cyber Threat Intelligence
What Is the Difference Between an IOC and an IOA?
An indicator of compromise (IOC) is a fixed piece of evidence tied to a known attack. Examples are a file hash, an IP address, a domain, a file name or a registry key. An indicator of attack (IOA) is a behavior: a sequence of actions an attacker has to perform, whatever tool they use.
In short, an IOC says “this exact thing was seen in an attack”. An IOA says “someone is doing what attackers do”.
The standard definitions are less tidy than this shorthand. In NIST’s glossary, IOCs are “technical artifacts or observables that suggest that an attack is imminent or is currently underway or that a compromise may have already occurred”. So in NIST’s wording, an IOC can also point to an attack in progress.
NIST’s glossary has no entry for indicator of attack; we searched it on 10 November 2026. Behavior does appear in the standards, just under another name. The glossary of the US Committee on National Security Systems, CNSSI 4009, has a behavioral definition. NIST quotes it on its indicator page: “a pattern of relevant observable adversary activity in the operational cyber domain”.
In daily SOC work the split is simple: IOCs are things, IOAs are actions.
| Question | Indicator of compromise (IOC) | Indicator of attack (IOA) |
|---|---|---|
| What is it? | A known artifact left by an attack | A sequence of actions an attacker has to perform |
| Typical examples | File hash, IP address, domain, URL, file name, registry key | A web server process starting a shell, a process reading LSASS memory, security tools switched off |
| What question does it answer? | Has this known bad thing touched our systems? | Is someone acting like an attacker right now? |
| When does it help? | When the same artifact shows up again, often after the fact | While the actions run, if the behavior is logged |
| What defeats it? | A recompiled file, a new server or a new domain | Activity that looks like normal admin work, or missing logs |
| How is it shared? | As a STIX Indicator pattern, in feeds over TAXII or in MISP | As ATT&CK technique IDs, STIX Attack Patterns and detection rules such as Sigma |
| Pyramid of Pain level | Bottom: hash values, IP addresses, domain names | Top: tools and TTPs |
| Where is it applied? | Firewalls, mail gateways, SIEM matching, EDR blocklists | EDR behavior engines, SIEM correlation rules, threat hunts |
For IOC types, context fields and a triage routine, read our indicator of compromise glossary entry and the IOC triage workflow. Behavior is described in the language of tactics, techniques and procedures (TTPs), which MITRE ATT&CK catalogues.
What Does an Indicator of Attack Look Like in Practice?
An IOA is an action plus its context. One event rarely proves anything, so a useful IOA names the process, its parent process, the account and what came before or after. The examples below come from public incident reports, with the ATT&CK technique each one maps to.
- A web server process starts a shell. Microsoft saw w3wp.exe, the SharePoint worker process, launch cmd.exe and encoded PowerShell in July 2025 (T1059.001, PowerShell).
- Something reads LSASS memory. In the same intrusions, Mimikatz ran its sekurlsa::logonpasswords module, which lists the credentials stored there (T1003.001, LSASS Memory).
- Security tools are switched off. The actor disabled Microsoft Defender through registry changes (T1685, Disable or Modify Tools).
- Group Policy pushes a script. The actor changed Group Policy Objects to spread ransomware across the domain (T1484.001).
- The Active Directory database is copied. ATT&CK lists the built-in ntdsutil.exe as one way to copy it (T1003.003, NTDS).
- Remote tools appear after an application is breached. Kaspersky names new remote-management or forensic tools after an application compromise as a runtime signal.
Each of these can also be legitimate. Admins run PowerShell, and IT teams edit Group Policy. That is why IOA rules carry false positive notes, and why joint guidance from CISA and partner agencies tells defenders to compare activity with a baseline of normal behavior.
How Does One Real Intrusion Look as IOCs and as IOAs?
In July 2025, attackers exploited on-premises Microsoft SharePoint servers through the flaws known as ToolShell, including CVE-2025-53770. Microsoft says exploitation attempts began as early as 7 July 2025. One actor, which Microsoft tracks as Storm-2603 and assesses with moderate confidence to be China-based, used the access to deploy Warlock ransomware from 18 July.
Microsoft’s report of 22 July 2025 lists both the artifacts and the behaviors of that chain, which makes it a clean test case. Our SharePoint ToolShell incident record covers the wider campaign. Here is the Storm-2603 chain, written as behavior.
- Exploit the serverA crafted request to an internet-facing on-premises SharePoint serverTechnique T1190 (MITRE ATT&CK, opens in a new tab)
- Plant a web shellspinstall0.aspx written to the LAYOUTS folderTechnique T1505.003 (MITRE ATT&CK, opens in a new tab)
- Run commands from the web serverw3wp.exe starts cmd.exe and encoded PowerShell, then whoamiTechnique T1059.001 (MITRE ATT&CK, opens in a new tab)
- Switch off protectionMicrosoft Defender disabled through registry editsTechnique T1685 (MITRE ATT&CK, opens in a new tab)
- Steal credentialsMimikatz reads LSASS memoryTechnique T1003.001 (MITRE ATT&CK, opens in a new tab)
- Push ransomware by policyGroup Policy changed to distribute WarlockTechnique T1484.001 (MITRE ATT&CK, opens in a new tab)
- EncryptFiles encrypted in the victim networkTechnique T1486 (MITRE ATT&CK, opens in a new tab)
The technique IDs follow Microsoft’s table, updated to ATT&CK v20.1. There, disabling security tools is T1685; Microsoft listed it as T1562.001 in July 2025. Now set the same chain against the indicators Microsoft published.
| Stage | As IOCs (Microsoft's list) | As IOAs (the behavior) |
|---|---|---|
| Initial access | Exploiting IPs such as 134.199.202[.]205, 104.238.159[.]149 and 188.130.206[.]168 | A crafted request to SharePoint followed by a new script file on the server |
| Web shell | spinstall0.aspx and variants such as spinstall1.aspx; SHA-256 92bb4ddb…a514 | The w3wp.exe worker process writing an .aspx file into the LAYOUTS folder |
| Commands | Hashes of four more web shells that call update[.]updatemicfosoft[.]com | w3wp.exe starting cmd.exe, then PowerShell with an encoded command, then whoami |
| Persistence | IIS_Server_dll.dll backdoor, listed with 12 SHA-256 hashes | The IIS worker loading a suspicious .NET assembly; a new scheduled task |
| Tunnel to the attacker | xd.exe, a fast reverse proxy, calling 65.38.121[.]198 | A web server host opening a new outbound proxy connection |
| Credential theft | No hash in Microsoft’s list | A process reading LSASS memory with Mimikatz |
| Defense evasion | No hash in Microsoft’s list | Defender protections switched off through registry edits |
| Spread and ransomware | No hash in Microsoft’s list | PsExec and Impacket running commands over WMI, then a GPO change that pushes Warlock |
The IOC column is precise and ready to use: anyone running SharePoint could sweep for those hashes and addresses as soon as Microsoft published them. But three of the eight stages have no artifact in the list at all. For those steps the actor used Mimikatz, registry edits, PsExec, Impacket and Group Policy. Microsoft described them as behaviors and published no hashes for them.
What Happens When the IOCs Are Not There?
On 5 November 2026, Kaspersky’s Securelist described a later intrusion that it attributes to Storm-2603 with moderate confidence. A single probe hit a SharePoint server on 20 December 2025, and the full exploit chain ran four days later. This time the web shell never touched the disk. It lived inside the w3wp.exe process and was lost when the application pool recycled or IIS restarted.
So a sweep for the July file names and hashes had nothing to match in that first stage. The attacking address, 45.91.94[.]204, was a node of the Safing Privacy Network, which routes traffic through several relays.
What stayed visible was behavior. The w3wp.exe process launched msiexec.exe to install Velociraptor, and LSASS was coerced into granting SYSTEM rights. New tunnels then opened through Zoho Assist, VS Code and Cloudflare Tunnel.
The attack was stopped late, at the ransomware stage. The actor dropped a vulnerable driver and a DLL that Kaspersky identified as Warlock’s AV killer, then tried to start the ransomware. Kaspersky says the Proactive Defense Module in its endpoint product blocked the ransomware process. Its conclusion is plain: detection “needs to move beyond file-based webshell hunting toward runtime behavior”.
Then the loop closed: Kaspersky published new IOCs from the case, including the attacking address, a hash for the web shell implant and the driver rsndispot.sys. The driver’s MD5 is 054a32d6033b1744dca7f49b2e466ea2. Other teams can now match what this one found through behavior. Our step-by-step analysis of the fileless Storm-2603 intrusion walks through every stage.
How Do the Two Look as Detection Logic?
An IOC check is a lookup. This PowerShell command hashes the .aspx files in one of the two LAYOUTS paths from Microsoft’s hunting query. It then compares them with the spinstall0.aspx hash Microsoft published.
Get-ChildItem "C:\Program Files\Common Files\microsoft shared\Web Server Extensions\16\TEMPLATE\LAYOUTS" -Recurse -Filter *.aspx | Get-FileHash -Algorithm SHA256 | Where-Object Hash -eq "92BB4DDB98EEAF11FC15BB32E71D0A63256A0ED826A03BA293CE3A8BF057A514"
A match is strong evidence, but an empty result proves little. Renaming a file keeps its hash, yet the smallest edit to its content changes it. An in-memory web shell leaves no file on disk for this sweep to hash.
An IOA check is a rule about behavior. The excerpt below is shortened from the public SigmaHQ rule “Suspicious Process By Web Server Process” (id 8202070f). It has fewer process names and no filters.
title: Suspicious Process By Web Server Process
tags:
- attack.t1505.003
- attack.t1190
logsource:
category: process_creation
product: windows
detection:
selection_webserver_image:
ParentImage|endswith:
- '\w3wp.exe'
- '\httpd.exe'
- '\nginx.exe'
- '\tomcat.exe'
selection_anomaly_children:
Image|endswith:
- '\cmd.exe'
- '\powershell.exe'
- '\pwsh.exe'
- '\certutil.exe'
- '\net.exe'
- '\reg.exe'
- '\wmic.exe'
condition: 1 of selection_webserver_* and selection_anomaly_children
falsepositives:
- Particular web applications may spawn a shell process legitimately
level: high The rule has no hash, no IP address and no web shell name. It fires on the July chain, where w3wp.exe started cmd.exe, and on any other web shell that makes the worker start a shell. Its own false positive note admits the cost: some web applications start shells for real reasons.
It also shows that IOAs need upkeep. In Kaspersky’s December case, the worker process started msiexec.exe, which the full SigmaHQ list does not include. A team that learns from that incident adds installers and remote-access tools to the rule.
Where Do IOCs and IOAs Sit on the Pyramid of Pain?
David Bianco’s Pyramid of Pain, first published in 2013, ranks indicator types by how much it hurts an attacker when defenders deny them. Hash values sit at the bottom and TTPs at the top. RFC 9424, an informational IETF document from August 2023, uses the same pyramid. It adds a second reading: the lower an indicator sits, the more fragile it is for the defender, and the more precise.
| Level | ToolShell example | What changing it costs the attacker |
|---|---|---|
| Hash values | 92bb4ddb…a514 for spinstall0.aspx | A trivial edit or a recompile gives a new hash |
| IP addresses | 134.199.202[.]205 | A new IP range or provider, sometimes a code change |
| Domain names | msupdate[.]updatemicfosoft[.]com | Choosing and registering a new domain |
| Network and host artifacts | spinstall0.aspx in the LAYOUTS folder; debug_dev.js holding MachineKey data | Reconfiguring or recompiling the tool |
| Tools | Mimikatz, PsExec, Impacket, Velociraptor | Finding or writing a new tool and learning it |
| TTPs | A web server process running a shell; credentials read from LSASS; ransomware pushed by Group Policy | Learning a new way to attack |
RFC 9424 warns against dropping the bottom of the pyramid. A fragile IOC is still worth deploying, because “there is no guarantee a fragile IoC will change”. The top is durable but costly: the RFC says TTPs and tools “take intensive effort to diagnose on the part of the defender”. That trade-off is IOC vs IOA in short.
Why Do IOCs Stop Working, and Why Do IOAs Fail Too?
IOCs age because attackers change them on purpose. CISA, the NSA, the FBI and partner agencies make the point in their 2024 joint guidance on living off the land. Threat actors, they say, “can easily modify known IOCs, such as filenames and command line arguments”. They also note “a general lack of conventional indicators of compromise” for attacks that use built-in tools.
The guidance gives an example. These two commands do the same thing, yet a string match on the first misses the second.
ntdsutil snapshot "activate instance ntds" create quit quit ntdsutil snapshot "ac i ntds" create quit quit
A behavior rule asks a different question: did ntdsutil create a snapshot of the Active Directory database on a domain controller? That holds for both spellings. The same thinking runs through our living off the land glossary entry and the Volt Typhoon profile. CISA says intrusions by that PRC state-sponsored group informed the guidance.
IOAs fail in other ways. They need telemetry such as process creation with command lines, and CISA lists default logging settings as one reason these attacks are hard to spot. They raise false positives, because admins use the same tools. And attackers go after the sensors themselves.
What Do Attackers Pay to Blind Behavior-Based Detection?
Criminal markets put a price on getting past behavior-based detection. Brandefense Blackhole, our cybercrime database, collected EDR killer offers between January and October 2026. The Cybercrime Report 2026 kept five dated offers. One asked $550 per build, another $950 for a driver-abuse build made to order, and a third USD 8,000 per slot for an exclusive kernel tool.
These are asking terms, not verified sales or tested tools. ESET, cited in the same analysis, tracked almost 90 EDR killers used in the wild, and 54 of them abuse a vulnerable but signed driver. In the December case, Kaspersky found a vulnerable driver, which it says many attackers use to terminate EDR tools, next to Warlock’s AV killer DLL. Read the full EDR killer price analysis for the offers and the defenses.
An EDR killer is itself a good IOC vs IOA test. The driver’s hash is an IOC, and Microsoft’s vulnerable driver blocklist blocks known drivers. The IOA is a new kernel driver loading on a server, followed by an endpoint agent that stops reporting. The EDR killer analysis suggests correlating missing agent heartbeats with host and network telemetry.
Some intrusions bring no malware at all. For the initial access chapter of the same report, we read 120 corporate network access listings in full. Counting one repeated firewall family once, 48 of 106 named an edge device or remote-access service.
When such access is sold as a working login, the buyer may leave no file to hash. The signals are then behavioral: an unusual VPN login, followed by discovery commands and admin tools. The chapter’s advice fits the IOC side too. Treat a list that names your domain “as a lead to check against your own logs, not as proof of a breach”.
Which Should a SOC Use, and When?
Neither replaces the other. The practical question is which one to reach for first in a given situation.
| Situation | Lean on | First step |
|---|---|---|
| An advisory publishes hashes, IPs and domains for an active campaign | IOCs | Sweep logs back to the campaign’s first known date, 7 July 2025 for ToolShell |
| A new exploit hits an internet-facing app and no IOCs exist yet | IOAs | Alert when the app’s server process starts shells, installers or remote tools |
| The attacker uses built-in tools such as PowerShell, WMI or ntdsutil | IOAs | Compare activity with a baseline of normal admin behavior |
| One host is confirmed compromised | IOCs | Extract its hashes, domains and IPs and sweep every other host |
| You share findings with peers or an ISAC | Both | Send IOCs with context and name the behaviors with ATT&CK technique IDs |
| Your domain appears in a dark web access listing | Neither yet | Treat it as a lead and check logins on the named service |
How Do IOCs and IOAs Work Together in One Workflow?
The two form a loop. Behavior finds the attack, artifacts spread the knowledge, and each incident should improve both lists.
- An IOA rule fires, for example w3wp.exe starting PowerShell with an encoded command.
- An analyst checks the surrounding events and confirms the activity is malicious.
- The team extracts the artifacts: hashes, file paths, domains, IP addresses and tool names.
- Those IOCs are swept across all hosts and back through stored logs to find the scope. A threat hunting pass looks for the same behavior elsewhere.
- The IOCs are shared with context. RFC 9424 names the threat actor, the IOC’s role in the attack, when it was last seen, its expected lifetime and related IOCs.
- Old IOCs are retired. RFC 9424 calls this end of life, for example after a shift in the actor’s TTPs or a remediation step.
- The IOA rule is tuned with the lesson, as with msiexec.exe in the Kaspersky case.
Sharing formats reflect the split. Here is the July web shell hash as a STIX 2.1 Indicator, trimmed (the id and timestamps a full object needs are left out). STIX and TAXII explains how such objects travel between tools.
{
"type": "indicator",
"spec_version": "2.1",
"indicator_types": ["malicious-activity"],
"name": "spinstall0.aspx web shell (ToolShell, July 2025)",
"pattern": "[file:hashes.'SHA-256' = '92bb4ddb98eeaf11fc15bb32e71d0a63256a0ed826a03ba293ce3a8bf057a514']",
"pattern_type": "stix",
"valid_from": "2025-07-22T00:00:00Z"
} The behavior side has its own STIX object, the Attack Pattern, which the standard calls a type of TTP. In practice, an IOA usually travels as an ATT&CK ID plus a rule. A hash can be matched anywhere, while a behavior rule depends on the logs each team collects.
What Should a Security Team Check First?
- List which IOC sources feed which control, and give every IOC a review or expiry date.
- Check that process creation events with command lines arrive from servers, not only from laptops.
- For each internet-facing application, alert when its server process starts shells, installers or remote tools.
- Map current rules to ATT&CK and look for empty stages, such as credential access or Group Policy changes.
- Alert when an endpoint agent stops reporting, not only when it raises an alert.
Brandefense Cyber Threat Intelligence delivers indicators with their context to SIEM, SOAR and EDR tools through its API and STIX/TAXII.
Also Called
- IOA vs IOC
- Indicators of compromise vs indicators of attack
Frequently Asked Questions
Is an IOA just a type of IOC?
It depends on whose definition you use. NIST's definition of an IOC already covers an attack that is currently underway, and RFC 9424 places TTPs at the top of its pyramid of IOC types. In daily use, security teams say IOC for fixed artifacts and IOA for behavior, and this page follows that usage.
Can the same evidence be both an IOC and an IOA?
Yes, depending on how you write it. The file spinstall0.aspx in the SharePoint LAYOUTS folder is an IOC. The w3wp.exe process writing any new .aspx file into that folder is an IOA, because it describes the action and would also catch a renamed web shell.
Can IOAs catch zero-day attacks?
They can, when the attack has to do something a rule watches for. A new exploit has no IOC yet, but running commands from a web server or reading LSASS memory can still match a behavior rule. An implant that stays inside a trusted process, as in Kaspersky's December 2025 case, may stay quiet until it starts new tools.
Are IOCs still worth using?
Yes. RFC 9424 says fragile IOCs are still worth deploying, because there is no guarantee the attacker will change them. IOCs are also the fastest way to scope an incident across many hosts and the easiest evidence to share with other teams.
How long should an IOC stay active?
There is no fixed period. RFC 9424 says how long an IOC stays useful depends on its confidence, fragility and precision, and that it should be removed at the end of its life to reduce false positives. Our IOC triage workflow sets review triggers for each control.
Sources
- Indicator of Compromise (glossary entry, from NIST SP 800-61r3), NIST Computer Security Resource Center1
- RFC 9424: Indicators of Compromise (IoCs) and Their Role in Attack Defence (August 2023), IETF2
- The Pyramid of Pain (David Bianco, 2013, updated 2014), Enterprise Detection & Response3
- Disrupting active exploitation of on-premises SharePoint vulnerabilities (22 July 2025, updated 23 July 2025), Microsoft Threat Intelligence4
- The SOC Files 5: Ghost in the pipeline, or a fileless SharePoint breach by Storm-2603 (5 November 2026), Kaspersky Securelist5
- Identifying and Mitigating Living Off the Land Techniques (joint guidance, 7 February 2024), CISA, NSA, FBI and partner agencies6
- STIX Version 2.1 (OASIS Standard), OASIS Cyber Threat Intelligence Technical Committee7
- Suspicious Process By Web Server Process (Sigma rule 8202070f), SigmaHQ8