- Learn more
- Threat Intelligence 101: What It Is and How It Works
- Product track
- Cyber Threat Intelligence
What Is the Pyramid of Pain?
The Pyramid of Pain is a model for choosing what to detect. It sorts the indicators of compromise a team can use by how much effort an attacker needs to replace each one after it is detected and blocked. Cheap indicators sit at the wide base. Costly ones sit at the narrow top.
David J. Bianco set it out on 1 March 2013 in a post on his blog, Enterprise Detection & Response. He wrote it after Mandiant released its APT1 report on 18 February 2013. Much of the discussion focused on the report’s long list of domain names, while its detailed data on the group’s tools got far less attention. His verdict was that “almost no one is using them effectively.”
His reasoning is short. You detect an indicator so that you can respond fast enough to deny the attacker its use. Denying some indicators costs the attacker nothing, while denying others forces real work. In Bianco’s words, “the amount of pain you cause an adversary depends on the types of indicators you are able to make use of”.
The model has six levels today. Bianco added the bottom one, hash values, in an update on 17 January 2014. The base is wide because indicators of those types are so plentiful, and the top is narrow.
What Are the Six Levels of the Pyramid of Pain?
Tough
Challenging
Annoying
Simple
Easy
Trivial
The figure shows the stack, with Bianco’s pain label under each level. The table below lists the levels from the bottom up and adds an example and the attacker’s next move.
| Level and pain | What it is | Example | Attacker’s next move |
|---|---|---|---|
| 1. Hash values Trivial | A fingerprint of one exact file, such as an MD5 or SHA-1 value. | The hash of one malware sample. | Changes one byte or rebuilds the file, which gives a new hash. |
| 2. IP addresses Easy | An address or netblock the attacker connects from or to. | A command and control (C2) server address. | Moves to another server, proxy or anonymity network. |
| 3. Domain names Simple | A domain or subdomain the attacker uses. | A C2 or phishing domain. | Registers and hosts a new domain, which costs a little money and time. |
| 4. Network and host artifacts Annoying | Traces that set the attacker’s activity apart from normal use. | A distinctive User-Agent string, URI pattern, registry key or service name. | Goes back to the lab to reconfigure or recompile the tool. |
| 5. Tools Challenging | Software the attacker brings to do the job. | A backdoor, a builder for spear-phishing documents or a password cracker. | Finds or writes a new tool, then learns to use it. |
| 6. Tactics, techniques and procedures (TTPs) Tough | How the attacker works, from reconnaissance to data theft. | Dumping cached credentials and reusing them in pass-the-hash attacks. | Learns new behavior, or gives up. |
The labels describe the attacker’s cost, not the indicator’s value to you. Bianco calls hashes “the most accurate type of indicator you could hope for”. They simply stop working the moment the file changes.
Why Do Hashes, IP Addresses and Domains Sit at the Bottom?
Any change to a file gives a new, unrelated hash. Bianco’s examples are “flipping a bit in an unused resource or adding a null to the end”. He places fuzzy hashes such as ssdeep higher, at the tools level, because they match similar files rather than one exact file.
Criminal markets sell that change as a service. The exploits and malware tooling chapter of our Cybercrime Report 2026 read a sample of 700 Brandefense Blackhole posts from 1 January to 27 October 2026. It found 29 crypter, loader and AV-bypass offers, and 18 of them claim to be FUD, short for “fully undetectable”. One crypting advert listed $50 per crypt or $400 a month for unlimited jobs, with the dollar signs as written.
Other sellers promised fresh stubs daily or capped buyers at two or five builds a day. We did not test these services, so these are advertised terms. Our study of what a malware subscription buys concludes that a tool found in one intrusion “may sit in a wrapper built for one buyer”. Its hash then says little about the next case.
Public code has the same effect. The UK’s NCSC says the Mimikatz source code is public, so anyone can compile their own version. A hash of one Mimikatz build will miss the next one.
IP addresses are the most basic indicator, Bianco notes, because an attack needs a network connection. A capable attacker can change them at will, for example through Tor or a similar anonymous proxy service. In May 2024, Mandiant reported that China-nexus espionage groups route traffic through ORB networks of hijacked routers and rented servers. An IPv4 address can stay in such a network for as few as 31 days, which Mandiant calls “IOC extinction”.
The KV botnet is a related case. The US Department of Justice says it turned old routers into relays that hid the origin of hacking it attributes to Volt Typhoon. A court-authorized operation disrupted it in December 2023.
Domain names cost a little more, since they must be registered, paid for and hosted. Some malware skips fixed domains entirely. MITRE ATT&CK describes domain generation algorithms (T1568.002) that compute fresh C2 domains, often by the hour, day or month. It notes that APT41 used them to change its C2 servers monthly.
None of this makes the bottom levels useless. A blocklist is cheap, precise and good for searching old logs during an incident. Give each value an end date, as our indicator of compromise triage guide explains, and do not count a long list as detection coverage.
What Do Network and Host Artifacts Look Like?
Artifacts are the traces that set an attacker’s activity apart from normal use. On the network, Bianco lists URI patterns, C2 data inside protocols and distinctive HTTP User-Agent or SMTP mailer values. On a host, they include registry keys a malware family creates, files dropped in set places and odd service names.
This is the first level where blocking costs the attacker real effort. Bianco’s example is a reconnaissance tool whose User-Agent string was off by one space or semicolon. Block that string and the attacker must work out how you caught it, then fix the tool.
Our threat library gives more examples. For Cobalt Strike, defenders hunt for known Beacon network patterns and named pipes. Persistence through registry run keys leaves entries you can compare with a known good baseline.
Why Do Tools and TTPs Hurt Attackers the Most?
At the tools level you take a whole program away. Bianco’s examples are antivirus or YARA rules that still match after moderate changes, fuzzy hashes and network tools with a distinctive protocol. The attacker must find or build a replacement and then learn to use it.
At the top you detect behavior, whatever tool performs it. Tactics, techniques and procedures describe how an attacker works. Bianco’s example is spotting pass-the-hash attacks in Windows logs rather than the tools behind them. MITRE ATT&CK lists pass the hash as T1550.002 and notes that APT1 used it.
Respond at this level across many TTPs, Bianco argues, and the attacker must either give up or reinvent itself from scratch. Covering every technique is a long job, so start by moving your most important detections up.
Stolen logins push defenders up the pyramid too. In our initial access offers chapter, 48 of 106 corporate access offers name an edge device or remote-access service, counting one repeated firewall family once. A buyer who signs in with valid accounts may bring no file to hash at all.
The same holds when attackers use tools already on the system, known as living off the land. In both cases, behavior is often the only clear signal: where the account signs in from, what it touches and in what order.
How Does One Detection Climb the Pyramid?
Take credential theft from LSASS memory, ATT&CK T1003.001 in our OS credential dumping entry. It is the job Mimikatz is best known for. You can catch the same activity at three levels.
- Hash level. Match the file hash of the Mimikatz binary from your last incident. A rebuilt copy slips past it.
- Tool level. MITRE’s analytic CAR-2019-04-004 flags processes that open lsass.exe with access masks and a call trace that Mimikatz uses. MITRE calls it “overtuned” to Mimikatz, and its Summiting the Pyramid team scores it 2U, a tool-specific detection.
- Behavior level. The same team dropped the call trace and kept two access masks, 0x1010 and 0x1410. In their tests most benign LSASS access used the mask 0x1fffff. The masks alone score level 4, and a filter that excludes the SYSTEM account brings the final analytic to 3U.
This is the original analytic from the Cyber Analytics Repository. It reads Sysmon Event ID 10 (process access) in Splunk:
index=__your_sysmon_data__ EventCode=10
TargetImage="C:\\WINDOWS\\system32\\lsass.exe"
(GrantedAccess=0x1410 OR GrantedAccess=0x1010 OR GrantedAccess=0x1438 OR GrantedAccess=0x143a OR GrantedAccess=0x1418)
CallTrace="C:\\windows\\SYSTEM32\\ntdll.dll+*|C:\\windows\\System32\\KERNELBASE.dll+20edd|UNKNOWN(*)"
| table _time hostname user SourceImage GrantedAccess
Below is the improved version, written by us in the same style from the fields MITRE scored. Test it on your own data before you trust it:
index=__your_sysmon_data__ EventCode=10
TargetImage="*lsass.exe"
(GrantedAccess=0x1010 OR GrantedAccess=0x1410)
NOT SourceUser="NT AUTHORITY\\SYSTEM"
| table _time hostname user SourceUser SourceImage GrantedAccess
Neither query is perfect. MITRE notes that an administrator may read LSASS memory in exactly the same way, so expect some false positives. The query also needs Sysmon process access events (Event ID 10) in your logs.
What Is MITRE’s Summiting the Pyramid?
Summiting the Pyramid is a project of the MITRE Center for Threat-Informed Defense. It turns the Pyramid of Pain into a way to score detection rules by how hard they are to evade. The documentation is at version 4.0.0. Its 2026 work adds a measure of detection coverage, a catalog of the ways each ATT&CK technique can be carried out and a Detection Coverage Calculator.
The model folds the first four pyramid levels into one row of ephemeral values. It splits tools and behaviors into two rows each. Columns show where the evidence comes from: the application, user mode or kernel mode on a host, or the header or payload of network traffic.
| Pyramid of Pain level | Summiting the Pyramid level (v4.0.0) | Name used on the LSASS example page |
|---|---|---|
| Hashes, IP addresses, domains, network and host artifacts | Level 1: Ephemeral | Ephemeral |
| Tools the attacker brings | Level 2: Implementation / Attacker-Controlled | Core to Adversary-brought Tool |
| Tools already on the system | Level 3: System-Constrained Interaction | Core to Pre-Existing Tool |
| TTPs, some ways of doing them | Level 4: Low-Variance Behaviors / Core Sometimes | Core to Part of (Sub-) Technique |
| TTPs, every way of doing them | Level 5: Invariant Behaviors / Core to Technique | Core to (Sub-) Technique |
A detection that relies on kernel events is harder to slip past than one built on application logs. On the network, evidence in the protocol payload is easier for an attacker to hide, for example through encryption, than evidence in the header. MITRE’s observables quick search gives a level 5 example: the registry key Windows creates when a task is scheduled, under Schedule\TaskCache. The project adds that this key can still fire on benign scheduled tasks.
Where Does Threat Intelligence Fit on the Pyramid?
A threat feed mostly fills the bottom three levels. Feeds carry hashes, addresses and domains that a SIEM or firewall can match at once. Reports on how a group works, and profiles mapped to MITRE ATT&CK, feed the top. The table runs from the bottom of the pyramid to the top and shows what tends to make each kind of intelligence go stale.
| What you receive | What makes it go stale | What to do with it |
|---|---|---|
| File hashes | A new build. Sellers in our Blackhole data advertise daily stubs and daily build caps. | Block, search old logs, set an end date. |
| IP addresses | Rotation. Mandiant saw ORB node addresses last as few as 31 days. | Block with an end date; check shared hosting before you block. |
| Domains | Re-registration. Many DGAs make a new domain every hour, day or month. | Block, then watch for the same registration pattern. |
| Malware analysis with artifacts and YARA rules | A rewrite or reconfiguration of the tool. | Turn into host and network detections. |
| Actor profiles with ATT&CK techniques | A change in how the group works, which is slow. | Write, test and hunt with behavior detections. |
Bianco’s test for any new report still works. For every paragraph, ask whether it gives you something to detect, and where that falls on the pyramid. Keep the hashes and domains, but plan your work around the parts higher up.
How Do Detection Teams Put the Pyramid of Pain to Work?
- Sort your current rules by levelCount how many alerts match a hash, IP or domain and how many describe behavior.
- Read every report against the pyramidNote each detail you can detect and the level it sits at.
- Keep cheap indicators, with an end dateBlock and search hashes, IPs and domains, then retire them on a schedule.
- Pick a few priority techniquesStart with the ATT&CK techniques your incidents and sector reports point to, such as credential dumping.
- Write and test one behavior rule for eachScore it with Summiting the Pyramid and run it against your own normal activity.
- Hunt where a rule is too noisy to alertRun the behavior as a threat hunting query on a schedule instead.
Rules kept as plain text are easier to score and share. A Sigma rule, for example, converts into a query for the SIEM you run. When a behavior is too common to alert on, use it in threat hunting, where an analyst reviews the results.
How Does the Pyramid of Pain Compare With the Kill Chain and ATT&CK?
| Model | Question it answers | Use it to |
|---|---|---|
| Pyramid of Pain (Bianco, 2013) | How much does detecting this hurt the attacker? It ranks six indicator types. | Choose what to detect and block. |
| Cyber kill chain (Lockheed Martin, 2011) | Which stage of the intrusion is this? It orders an attack into seven stages. | Put an incident in order and find stages with no defense. |
| MITRE ATT&CK | What exactly did the attacker do? It names tactics, techniques and procedures. | Give detections and reports shared names. |
| Summiting the Pyramid (MITRE) | How hard is my rule to evade? It scores rules on five levels and event columns. | Score and improve detection rules. |
The models work together. The kill chain tells you where an attack can be stopped, and ATT&CK names what happened. The pyramid tells you how much the attacker must change once you catch it.
What Are the Limits of the Pyramid of Pain?
- It measures the attacker’s cost, not yours. Behavior rules need the right logs, such as Sysmon process access events, and time to tune.
- Higher can be noisier. MITRE warns that a tool-level rule can alert on benign activity when the tool is built into the operating system.
- Low levels still pay off. A hash gives few false alarms, and old hashes and IPs help you scope an incident by searching past logs.
- Built-in tools blur the tools level. That is why Summiting the Pyramid separates tools the attacker brings from tools already on the system.
- Pain can be bought off. An actor who uses a contractor-run ORB network, or pays for a crypter subscription, hands the work of replacing blocked addresses and files to someone else.
- It is not a timeline. It does not say when an indicator appears in an attack. The kill chain does that.
Brandefense Cyber Threat Intelligence pairs indicators for SIEM hunting with actor profiles and the techniques they use, so a team can work at both ends of the pyramid.
Also Called
- David Bianco's Pyramid of Pain
Frequently Asked Questions
Who created the Pyramid of Pain?
David J. Bianco described it in a blog post on 1 March 2013, written after Mandiant's APT1 report. He updated it on 17 January 2014 and added hash values as a separate bottom level, which gave the model the six levels it has today.
What are the 6 levels of the Pyramid of Pain?
From the bottom up: hash values, IP addresses, domain names, network and host artifacts, tools, and tactics, techniques and procedures (TTPs). Bianco labels the pain to the attacker as trivial, easy, simple, annoying, challenging and tough.
Why are hash values at the bottom of the Pyramid of Pain?
Changing a single bit of a file gives a completely different hash, so an attacker can make a known hash useless with almost no effort. Crypter services advertised on criminal markets sell exactly that, with some promising fresh stubs every day.
What is the difference between the Pyramid of Pain and the cyber kill chain?
The cyber kill chain splits an intrusion into seven ordered stages, from reconnaissance to actions on objectives. The Pyramid of Pain ranks the indicators you can detect by how costly they are for the attacker to replace. The kill chain shows where an attack can be stopped, and the pyramid helps you choose what to detect there.
Is the Pyramid of Pain still relevant?
Yes. The MITRE Center for Threat-Informed Defense built its Summiting the Pyramid project on it and was still extending that project in 2026. Short-lived proxy networks and paid re-crypting make low-level indicators go stale faster, which strengthens the case for detecting behavior.
Sources
- The Pyramid of Pain (David J. Bianco, 1 March 2013, updated 17 January 2014), Enterprise Detection & Response blog1
- Summiting the Pyramid v4.0.0: Detection Engineering Work Overview, MITRE Center for Threat-Informed Defense2
- Summiting the Pyramid: LSASS Memory analytic, MITRE Center for Threat-Informed Defense3
- CAR-2019-04-004: Credential Dumping via Mimikatz, MITRE Cyber Analytics Repository4
- IOC Extinction? China-Nexus Cyber Espionage Actors Use ORB Networks to Raise Cost on Defenders (22 May 2024), Mandiant, Google Cloud5
- Dynamic Resolution: Domain Generation Algorithms (T1568.002), MITRE ATT&CK6
- Use Alternate Authentication Material: Pass the Hash (T1550.002), MITRE ATT&CK7
- Cybercrime Report 2026: Exploits, Malware Tooling and AI-Assisted Crime, Brandefense8