OT Cybersecurity
Diamond Model - a practical guide to incident analysis
The Diamond Model of Intrusion Analysis - a method for incident analysis: adversary, capability, infrastructure, victim. Pivoting, activity threads and OT examples.
Józef SulwińskiPublished Updated 10 min read

Contents
Blocking the IP address from an alert closes the ticket, but it does not say who is attacking, which other systems are at risk or whether this is part of a larger campaign. A security team that knows only an address and a file hash answers every alert from scratch.
The 2013 paper by Sergio Caltagirone, Andrew Pendergast and Christopher Betz shows that:
- every event on a network is described by four features: adversary, capability, infrastructure and victim,
- from one known feature an analyst can reach the others, and this step is called pivoting,
- events are linked into activity threads, and threads into activity groups.
The Diamond Model of Intrusion Analysis is therefore a method for turning a single indicator of compromise (IOC, that is an address, a hash or a domain) into a picture of a whole campaign.
The four vertices of the diamond
Every security incident is described by four elements connected in a diamond shape. For each, the analyst should record what is known and leave the field empty when knowledge is missing.
| Vertex | What it describes | What to establish | Example (Stuxnet) |
|---|---|---|---|
| Adversary | The person or organisation responsible for the attack | The motivation, the resources and whether the actor works for a state or commercially. | The sources used here do not name the actor. |
| Capability | Tools and techniques of the attack | Which malware and which vulnerabilities were used, and whether the tools are custom or purchased. | Multiple zero-day vulnerabilities, a rootkit in the PLC and modification of S7-315-2 controller code. The code for S7-417 was disabled or unfinished. |
| Infrastructure | Technical resources through which the attack runs | Which domains, servers and media delivered and controlled the attack. | Five organisations with a presence in Iran, USB drives and C2 servers. Symantec assumes that a third-party contractor introduced it. |
| Victim | The target of the attack | Which sector and which systems were attacked, and why this target. | According to an ISIS analysis the target was IR-1 centrifuges at Natanz. Iran has not admitted that Stuxnet attacked Natanz. |
Five edges of the diamond connect the vertices in pairs, and each pair is an analytical question:
- Adversary - capability: What tools has the adversary developed, and were they bought or built in-house.
- Adversary - infrastructure: What infrastructure does the adversary control, and are these its own servers or compromised ones.
- Infrastructure - capability: How the tools use the infrastructure for communication.
- Infrastructure - victim: Through what infrastructure the attack on the victim ran.
- Capability - victim: Which techniques worked and how the attack got in.
The extended model describes a sixth relationship, adversary-victim, as a social-political meta-feature. The adversary rarely acts directly and usually works through infrastructure, so this relationship is a hypothesis to test and not an edge of the event graph.
TIP
For every incident, the diamond should be filled in: who (adversary), with what (capability), through what (infrastructure), whom (victim). Why: empty fields show what the team does not yet know and point to where further analysis should start.
Pivoting - the power of the Diamond Model
Pivoting means moving from one known element to related, unknown ones. The authors call it one of the model’s strongest characteristics, and the diamond itself emerged from such pivot scenarios.
Example: a team finds malware on the OT network (it knows the capability) that connects to an IP address (it knows the infrastructure). From that point it should establish four things:
- Infrastructure to adversary: who controls this address and whether it appears in threat reports. Why: a known adversary has documented methods that point to the next steps of the attack.
- Infrastructure to capability: what other tools connect to this address and whether the network holds more malware. Why: one address often serves several samples.
- Adversary to victim: which other organisations this adversary attacked and whether it targets a whole sector. Why: the answer shows whether the company is a target or a chance victim.
- Capability to victim: which systems on the network may be vulnerable to the same techniques. Why: these systems should be checked before the attack repeats.
Each pivot generates new questions and new indicators to check, so a single alert becomes a map of connections.
TIP
Pivoting uses the data at hand: sample analysis, domain name resolution, firewall logs and IP address registries. Threat information sharing platforms widen the view, but they are not a precondition for starting.
Activity threads and activity groups
The model defines two higher levels of analysis.
Activity thread
An activity thread is the phase-ordered events of one adversary against one victim, linked causally. It describes one pass of the attacker through the Kill Chain:
E1 (reconnaissance) -> E2 (delivery) -> E3 (exploitation) -> E4 (C2) -> E5 (exfiltration)
Each event is a separate diamond with four vertices. An arrow between events means that the earlier event enabled the next one.
Activity group
An activity group collects events and threads with similar features. It usually serves to point to a common adversary. A shared public capability or shared hosting does not suffice to merge events into one group, because the authors warn against overfitting.
| Level | What it describes | Example |
|---|---|---|
| Event | A single event with four vertices. | Malware connects to a C2 server. |
| Activity thread | Events of one attack, linked causally. | A phishing message delivers a program, the program connects to a C2 server and then exfiltrates data. |
| Activity group | Events and threads with shared features. | The Sandworm campaign against Ukrainian power grids in 2022 (MITRE G0034). |
Practical analysis - two OT incidents
Two incidents, Colonial Pipeline and Industroyer, show how the diamond changes conclusions.
| Vertex | Colonial Pipeline (2021) | Industroyer (2016) |
|---|---|---|
| Adversary | The DarkSide group, which operated a ransomware-as-a-service (RaaS) model with a financial motive. | Sandworm Team. A US indictment of October 2020 names GRU officers of unit 74455, but an indictment is not a verdict. |
| Capability | Ransomware with data theft, and access through a VPN profile that was not intended to be in use. | A framework with four OT protocol modules: IEC 101, IEC 104, IEC 61850 and OPC DA. |
| Infrastructure | One VPN profile with single-factor authentication, and according to CISA DarkSide primarily uses Tor for C2. | C2 servers mostly in the Tor network. |
| Victim | A fuel pipeline operator whose IT network the ransomware encrypted. | The Pivnichna substation of the operator Ukrenergo, specifically the substation’s control systems. |
| What the diamond shows | A financial attack on the IT network. The company shut the pipeline down as a precaution, and according to CISA there are no indications of lateral movement to OT. | An attack on critical infrastructure with a tool that controls devices in OT protocols. |
| Defence | The IT network and the dependencies of the OT process on it should be secured: multi-factor authentication, backups, an incident response plan and IT-OT segmentation. | The OT network should be segmented, industrial protocols monitored and manual operating procedures prepared. |
Analysis of the adversary vertex changes the defence strategy. For a financial attack the IT network and the OT process’s dependencies on it should be secured first. Colonial stopped the pipeline although according to CISA OT was not attacked, because the company’s chief executive told the Senate that even a one percent chance of an OT compromise justified the shutdown. For a tool like Industroyer the defence must cover the OT layer, because the malware controls circuit breakers directly in industrial protocols.
The Diamond Model and other methods
The three methods complement each other. The authors of the model write that the Cyber Kill Chain is a highly effective and influential model and that the Diamond Model complements it.
| Method | Answers the question | What it adds |
|---|---|---|
| Cyber Kill Chain | At what stage is the attack? | Orders the attack phases and helps plan layered defence. |
| MITRE ATT&CK | What technique did the attacker use? | Describes techniques. Group and campaign pages hold context about the adversary, but the model is not meant for analysing the adversary-victim relationship. |
| Diamond Model | Who is attacking, with what, and whom? | Links the features of an event into relationships and lets the analyst move between them. |
An OT incident should be analysed from three angles:
- Kill Chain: establish at which stage the attack was detected and where the defence failed.
- MITRE ATT&CK: establish which techniques were used and which detections to add.
- Diamond Model: establish who is behind the attack, whether further attacks are likely and what the defence strategy should be.
How to implement the Diamond Model in an organisation
TIP
To start, a diamond template in the form of a simple table with four vertices, filled in for each incident, is enough. Why: after a few reports the tables show recurring infrastructures, tools and targets.
Implementation checklist
- A diamond template (four vertices and five edges) should be created in the incident management tool. Why: one template gives comparable descriptions.
- The diamond should be filled in for every incident, even when not all vertices are known. Why: empty fields show what to investigate next.
- Indicators of compromise should be collected and assigned to vertices (an IP address to infrastructure, a file hash to capability). Why: an indicator assigned to a vertex shows where to pivot from it.
- The analyst should pivot from every known vertex to unknown ones. Why: this leads from a single alert to the full picture.
- Events should be linked into activity threads causally, in the order of phases. Why: chronological order alone does not prove that events are related.
- Threads with shared features should be grouped into activity groups, and the analyst should check that the shared feature is really specific to the adversary. Why: a shared public capability or shared hosting leads to wrong groups.
- The team should move to a threat information sharing platform when the number of incidents outgrows the tables. Why: a platform automates the search for links.
- Analyses should be shared with the ISAC for the sector (an information sharing and analysis centre; ENISA maintains the list). Why: other members can fill in the vertices the organisation does not know.
Model extensions - meta-features
The paper defines additional meta-features for each event. Meta-features order events and make them comparable.
| Meta-feature | Description | Example |
|---|---|---|
| Timestamp | The date or date range of the event. | 17-18 December 2016 (attack on the Pivnichna substation, according to CCDCOE). |
| Phase | The attack phase according to the chosen model. One example should use one model. | Delivery (a Cyber Kill Chain phase). |
| Result | Success, failure or unknown. | Success. |
| Direction | One of seven values, among them infrastructure to victim, victim to infrastructure or bidirectional. | Infrastructure to victim. |
| Methodology | The general class of activity. | A spear-phishing email with an attachment. |
| Resources | External elements the event depends on: software, knowledge, hardware, funds and access. | Public malware, internet access and funds to buy domains. |
Sources
- The Diamond Model of Intrusion Analysis - Caltagirone, Pendergast, Betz (Center for Cyber Intelligence Analysis and Threat Research); copy on the authors' site, because the DTIC address (ADA586960) is currently unavailable (accessed: )
- W32.Stuxnet Dossier (version 1.4, February 2011) - Symantec (accessed: )
- Did Stuxnet Take Out 1,000 Centrifuges at the Natanz Enrichment Plant? (22 December 2010) - Institute for Science and International Security (Albright, Brannan, Walrond) (accessed: )
- Stuxnet, Software S0603 - MITRE ATT&CK (accessed: )
- Alert (AA21-131A): DarkSide Ransomware: Best Practices for Preventing Business Disruption from Ransomware Attacks - CISA (accessed: )
- Testimony of J. Blount before the US Senate, 8 June 2021 - U.S. Senate / GovInfo (accessed: )
- Win32/Industroyer: A new threat for industrial control systems - ESET (accessed: )
- Industroyer - Crash Override (2016) - CCDCOE Cyber Law Toolkit (accessed: )
- Sandworm Team, Group G0034 - MITRE ATT&CK (accessed: )
- Six Russian GRU Officers Charged in Connection with Worldwide Deployment of Destructive Malware and Other Disruptive Actions in Cyberspace - U.S. Department of Justice (19 October 2020) (accessed: )
- Information Sharing and Analysis Centers (ISACs) - ENISA (accessed: )
Need help in this area?
Our experts will help you assess the risk and plan next steps.