In 2011, three analysts at Lockheed Martin, Eric Hutchins, Michael Cloppert and Rohan Amin, published a paper on what they called intrusion kill chains. The idea was borrowed from military targeting: an attack is not a single event but a sequence of steps, and if the defender breaks any one link, the attacker has to start again. That framing changed how many security teams think. Instead of asking “were we breached?”, they started asking “how far did the attacker get, and where could we have stopped them earlier?”
The model has seven stages. The easiest way to understand them is to follow one imagined intrusion all the way through.
One intrusion, seven stages
Picture a mid-sized logistics company in Pune with about 800 employees. A criminal group wants to deploy ransomware there.
1. Reconnaissance. The group studies the company from the outside. LinkedIn shows who works in accounts payable. The company website lists a vendor portal. Job postings mention the VPN product and the email platform in use. None of this touches the company’s systems, which is exactly why it is hard to see. At most, defenders might notice unusual scanning of their internet-facing services.
2. Weaponisation. The attacker prepares the tool: perhaps an invoice-themed document carrying a malicious macro, or a link to a fake login page. This happens entirely on the attacker’s side. Defenders cannot observe it directly, but threat intelligence about the group’s favourite lures and file types can be used to tune email filtering in advance.
3. Delivery. The weapon reaches a target. Email is the classic route, but delivery can also be a compromised website, a USB drive, or an exposed remote access service. This is the first stage where the company’s own controls get a direct look: the email gateway, web proxy and attachment sandbox all operate here.
4. Exploitation. Something executes. A user enables content in the document, or a software flaw is triggered. Patching, macro restrictions and exploit protection on the endpoint matter most at this point.
5. Installation. The attacker establishes a foothold that survives a reboot: a scheduled task, a new service, a registry entry that launches a program at logon. Endpoint detection tools watch closely for these persistence changes.
6. Command and control. The implant reaches out to infrastructure the attacker controls and waits for instructions. Network monitoring, DNS logs and proxy logs are the main sources here, looking for regular, automated-looking connections to rare destinations.
7. Actions on objectives. The attacker does what they came for. In our example that means moving through the network, stealing a copy of the finance share for extortion, deleting backups and finally encrypting servers. Everything earlier was preparation for this.
The real point: earlier is cheaper
The value of the chain is not the labels. It is the observation that a defence at stage 3 costs far less than a response at stage 7. Blocking the email costs one quarantine action. Catching the implant at stage 6 means reimaging a laptop and resetting a password. Discovering the attack at stage 7 can mean days of outage, a regulatory report and a ransom demand.
The original paper paired the stages with a “courses of action” matrix. For each stage, defenders ask whether they can detect, deny, disrupt, degrade, deceive or destroy. Filling that grid honestly is a useful exercise. Many organisations find their matrix is heavily weighted towards detection at delivery (email filtering) and almost empty at installation and command and control.
A second idea from the paper gets less attention but is just as useful: when an attack is stopped, analysts should still reconstruct the stages that happened before the block. If the email gateway caught a lure, what did reconnaissance look like? Which sender infrastructure was used? Those details, collected across many blocked attempts, reveal the campaign behind them and let defenders anticipate the next variation.
Using it during an investigation
When an alert fires, placing it on the chain gives the analyst two immediate questions. Looking backwards: how did this get here? An alert for a suspicious scheduled task (installation) implies delivery and exploitation already happened, so the analyst should search mail and browsing logs for the entry point. Looking forwards: what comes next? If installation succeeded, the analyst should check for command and control traffic and early signs of lateral movement.
This simple “backwards and forwards” habit prevents a common mistake, which is closing an alert after cleaning the one artefact that triggered it. Deleting a malicious scheduled task does nothing about the stolen credentials the attacker may already be using elsewhere.
Where the model falls short
The kill chain was built around a specific kind of attack: an outsider delivering malware to break into a perimeter network. Several modern patterns do not fit it neatly.
- Credential-based attacks. If an attacker logs in to a cloud email account with a stolen password, there is no weaponisation, exploitation or installation in the traditional sense. The chain jumps from reconnaissance almost straight to actions on objectives.
- Insiders. An employee copying customer data to a personal drive skips nearly every stage.
- Everything after the foothold. Lateral movement, privilege escalation, discovery and data collection all get squeezed into stage 7, even though in a real intrusion that is where most of the attacker’s time and the defender’s best detection opportunities sit.
- Perimeter bias. The model implicitly encourages investment at the edge. Attackers who get past the edge, which some always will, meet less resistance.
Real attacks also loop. An attacker inside one machine performs fresh reconnaissance of the internal network, delivers tools to a second machine and installs persistence there. The chain is better read as a cycle that can repeat at each hop than as a single straight line.
Kill chain and ATT&CK together
These gaps are one reason MITRE ATT&CK became popular. ATT&CK breaks the post-compromise phase into many tactics, such as persistence, privilege escalation, credential access, lateral movement and exfiltration, and lists specific techniques under each. Other variants exist too, for example the Unified Kill Chain, which merges ideas from both into a longer sequence.
In practice the two models do different jobs. The kill chain is a good communication tool: a manager instantly understands “we stopped it at delivery” versus “it reached actions on objectives”. ATT&CK is a working vocabulary for analysts and detection engineers who need precision about what exactly the attacker did. Many SOC reports use the kill chain stage as the headline and ATT&CK technique IDs in the detail.
Used with that understanding, the kill chain remains worth knowing. It reminds defenders that attacks take time, that each step leaves traces, and that the cheapest place to win is always earlier than where you are now.
Written by the RP Strativa Team. The company in the example is illustrative.
Leave a Reply