OT Cybersecurity
Cyber Kill Chain - a model for analyzing cyberattacks
Lockheed Martin's Cyber Kill Chain - seven phases of a network intrusion. How to use the model for defense planning and how it relates to MITRE ATT&CK and the ICS Cyber Kill Chain.
Józef SulwińskiPublished Updated 7 min read

Contents
A security team sees alerts from individual systems and does not know at what stage an attack is or where to stop it. The Cyber Kill Chain model sorts such alerts into seven phases of a network intrusion.
The paper by three Lockheed Martin analysts, Eric Hutchins, Michael Cloppert and Rohan Amin, published in 2010-2011, shows that:
- The paper calls the model the intrusion kill chain. Lockheed Martin also uses the name Cyber Kill Chain, which is a registered mark.
- The term kill chain comes from military doctrine and means a sequence of steps leading to neutralizing a target. The absence of one link breaks the whole process.
- The model describes network intrusions in which a payload crosses the victim’s network boundary, for example in campaigns by APT groups, that is, advanced, long-running intrusion campaigns.
- According to the authors, defense should be moved to earlier phases.
The seven phases of the Cyber Kill Chain
The paper describes seven phases. The authors also present a matrix of courses of action: a defender can detect, deny, disrupt, degrade, deceive or destroy an event.
| Phase | Name | What the attacker does | How a defender can break it |
|---|---|---|---|
| 1 | Reconnaissance | Gathers information about the target: email addresses, social relationships and technologies in use. | Limits public information about the organization and watches publicly available sources (OSINT). |
| 2 | Weaponization | Combines an exploit with a payload, for example in a PDF or Office document. | Analyzes threat intelligence and tracks campaigns. |
| 3 | Delivery | Delivers the tool to the victim through an email attachment, a website or a USB drive. | Filters email, runs suspicious files in an isolated environment (a sandbox) and trains users. |
| 4 | Exploitation | Exploits a vulnerability to run the attacker’s code. | Installs patches, hardens configuration and limits program execution to an allowlist. |
| 5 | Installation | Installs persistent access, for example a backdoor. | Uses endpoint protection (EDR), which detects suspicious processes, and an application allowlist. |
| 6 | Command and Control | Sets up a communication channel with the attacker’s infrastructure. | Monitors outbound traffic, filters DNS and uses a network detection system (NDR). |
| 7 | Actions on Objectives | Achieves the goal: steals data, encrypts systems or manipulates a process. | Uses data loss prevention (DLP), segmentation, anomaly monitoring and backups. |
Application in practice
A defender does not have to stop the attacker at every stage, because the model assumes that every link is needed. If a phishing message gets past the email filter (phase 3), a patch can block the exploit (phase 4). If the exploit works, endpoint protection can detect the installation of a backdoor (phase 5).
Example: Colonial Pipeline in the model
In May 2021, DarkSide ransomware encrypted the IT systems of Colonial Pipeline. The company’s CEO testified to the US Senate on 8 June 2021 that the attacker exploited a legacy VPN profile that had only single-factor authentication. In June 2021, the sources did not say how the attacker obtained the password.
According to CISA, there was no indication of a direct impact of the attack on OT, and the company proactively disconnected some OT systems. The CEO testified that the pipeline was shut down because compromise of OT could not be ruled out.
Reading the incident in the model is an interpretation, not a finding of the sources.
| Phases | How the incident can be read in the model |
|---|---|
| 1-4 | Entering through a VPN account bypasses weaponization, delivery and exploitation in the paper’s sense, because the sources describe no payload or exploit. This is a limit of a model that assumes a payload. |
| 5-6 | The sources describe neither the installation of persistent access nor a command channel. |
| 7 | The ransomware encrypted IT systems, and the operator shut down the pipeline. |
How to start
- Assessment of controls. The head of the security team maps controls to the seven phases and records which tool or process covers each phase. The result is a coverage table. A phase without coverage should go into the action plan, because an attacker can pass it unhindered.
- Incident analysis. An analyst assigns events to phases. If there are no traces for a phase, it should be recorded that the phase may have been skipped or not recorded, because this points to a detection gap or a limit of the model. The analysis should then be extended with MITRE ATT&CK techniques.
- Report for management. The report should be built on the seven phases, with technical details attached as an annex with ATT&CK techniques. Seven phases are easier to discuss than 15 Enterprise tactics.
- Priorities. The Kill Chain describes the course of an attack. Priorities follow from a risk assessment that takes the industry, assets and threat profile into account. Data on groups active in a given sector comes from MITRE ATT&CK.
Limits of the model
The model has five limits:
- Scope. The model describes intrusions in which a payload crosses the network boundary. It maps attacks without a payload, for example with stolen credentials, only in part.
- Order. The model assumes consecutive phases. An attack with a stolen account skips some of them.
- Insider threats. The model describes an attack from outside. Separate phase models exist for insider threats, for example the one by Duran and co-authors from 2009, listed in the paper’s bibliography.
- Supply chain and cloud. In the author’s assessment, attacks through a supplier, and cloud misconfiguration and token theft, are hard to map onto seven phases. The attacker does not send a payload across the victim’s network boundary.
- AI agents as an attacker’s tool. Anthropic described an espionage campaign detected in September 2025 in which an AI agent performed 80-90% of the actions: reconnaissance, exploit code, credential collection and exfiltration. A human made 4-6 decisions per campaign. The campaign covered about 30 targets, and the intrusion succeeded in a small number of cases, because the agent was unreliable and invented credentials. Anthropic attributed the campaign to a Chinese state-sponsored group with high confidence.
The model works as a map for talking to management and a starting point for analysis in ATT&CK.
Cyber Kill Chain compared with MITRE ATT&CK and the ICS Cyber Kill Chain
ATT&CK describes in detail what the seven phases of the Kill Chain do not cover.
| Aspect | Cyber Kill Chain | MITRE ATT&CK |
|---|---|---|
| Level of detail | The model has seven phases of an intrusion. | In version 20 (from 13 October 2026) the Enterprise matrix has 15 tactics, 227 techniques and 479 sub-techniques. |
| Order | The model assumes an order of phases. | Techniques are grouped into tactics, and the order in which they are used depends on the attack. |
| Link to adversaries | The model does not tie phases to specific groups and software. The paper gives a courses-of-action matrix. | ATT&CK ties techniques to groups, software and adversary tools. |
| Detection | The paper gives general courses of action for each phase. | Version 20 has 703 Detection Strategies and 1770 Analytics. |
| Industrial systems | The model does not distinguish IT from OT networks. Assante and Lee described an adaptation to industrial systems in the ICS Cyber Kill Chain (SANS, 5 October 2015), which has two stages. | A separate ICS matrix has twelve tactics, 78 techniques and 26 sub-techniques in version 20. |
The two models complement each other. The Kill Chain shows at what stage an attack is. ATT&CK shows which technique the attacker used and how to detect it. The ATT&CK Navigator makes it easier to map controls onto the ATT&CK matrix.
Where we help
Penetration tests show what an attacker achieves through weaknesses. Red teaming checks whether the defense team detects and stops a multi-stage attack on the organization.
Sources
- Intelligence-Driven Computer Network Defense Informed by Analysis of Adversary Campaigns and Intrusion Kill Chains - Lockheed Martin (accessed: )
- The Cyber Kill Chain - Lockheed Martin (accessed: )
- DarkSide Ransomware: Best Practices for Preventing Business Disruption from Ransomware Attacks (AA21-131A) - CISA (accessed: )
- Hearing before the Senate Committee on Homeland Security and Governmental Affairs, 8 June 2021 (testimony of J. Blount) - U.S. Government Publishing Office (accessed: )
- Colonial Pipeline CEO: Ransomware Attack Started via Pilfered 'Legacy' VPN Account - Dark Reading (accessed: )
- Updates - October 2026 (ATT&CK v20) - MITRE ATT&CK (accessed: )
- The Industrial Control System Cyber Kill Chain - SANS Institute (accessed: )
- Disrupting the first reported AI-orchestrated cyber espionage campaign - Anthropic (accessed: )
Need help in this area?
Our experts will help you assess the risk and plan next steps.