There are few topics in cybersecurity that inspire debate quite like attribution. For exactly that reason, I’m excited to have the chance to bring people together to have a discussion about it at the upcoming RSA conference in an all-new format known as a cooperative learning session. Here’s a preview of what we’ll discuss in the session, including some basics of attribution and what makes it so difficult.
A good place to start in order to have a productive discussion is by defining the word “attribution” in a cybersecurity context. A simple definition I use is that attribution is the act of associating cyber activity with something else. I admit, that’s a frustrating and broad definition, but I use it for a reason: many people define and perform attribution differently, which is what leads to this being such a tough topic for us to discuss.
Here are a few common types of attribution that analysts commonly use. I like to bin these into the who and the how—because as we’ll discuss next, sometimes one of these types matters more than another.
These types of attribution focus more on who is behind the activity. Analysts might attribute activity to:
- A person/persona: The person behind the keyboard who performed an intrusion or activity developed the malware in question. Analysts may start tracking this via a persona (sometimes based on a handle or account name) and eventually identify the person behind that persona.
- A team, unit, or organization: The group of people behind the activity, whether they are a loosely formed hacktivist group, an organized military unit, or a company’s red team.
- A government: The country behind the activity. (This can get even more complex, because “state-sponsored” isn’t straightforward, as Jason Healey discusses in this paper.)
Analysts can also perform attribution based on how activity happened, which may completely ignore the who behind the keyboard. When attributing based on how, analysts often look for some unique attributes of the activity, such as a unique domain, execution sequence, command-line options, code snippet, or some combination of the above.
Analysts might attribute activity based on:
- Tools/malware: Adversaries use various tools and malware, both open and closed source, throughout their intrusions. Sometimes these are custom and unique to adversaries, but oftentimes different adversaries use the same tools.
- Other code: Adversaries use other code and scriptlets, such as PowerShell, throughout their intrusions.
- Tactics, techniques, and procedures (TTP): More broadly, adversaries use TTPs to achieve their goals. MITRE ATT&CK provides a common framework of TTPs that analysts can use to track adversary behaviors.
- Infrastructure: Adversaries register and maintain domains, Internet protocol (IP) addresses, and other infrastructure, which can provide unique pivot points for analysts.
Why not both?
Of course, analysts can also combine the who with the how and make attribution assessments based on a mix of the two.
Separating the who from the how is inspired by (and can be nicely summarized in) the Diamond Model. It divides the who into the Social-Political Axis, which also includes the victim in addition to the adversary, and the how into the Technology Axis, consisting of capability and infrastructure. As the paper describes, when using the Diamond Model to create activity groups, analysts can choose from the four features, allowing them to focus on the who, the how, or both.
To DIY or not?
To make attribution even more complicated, we have another choice as we do it: we can either create our own clusters to attribute to, or we can attribute to clusters that other teams have created.
Analysts can attribute to:
- A cluster of activity they observe: Analysts might also take a look at activity and intrusions they have visibility on and decide to cluster them based on a methodology they choose. There are many different ways to cluster activity—including campaigns, intrusion sets, activity groups, and threat groups—and each team chooses what works for them. Sometimes these clusters are associated with just the who, just the how, or both the who and the how. One method of clustering that allows analysts to choose whether they cluster based on the who or the how is by creating activity groups based on the Diamond Model. Activity groups may ignore the Adversary feature altogether and focus on clustering based on features like Capability (e.g., malware or techniques) or Infrastructure (e.g., command and control domains or IPs).
- A cluster of activity someone else has named: When analysts perform attribution, they might associate what they’re seeing with a name that another team has created. For example, FireEye uses the UNC, FIN, and APT designations, and a team might note that activity they’re observing overlaps with activity from a group like FIN7.
When does attribution matter?
That brings us to the crux of this debate: does attribution really matter? Well, it depends on both how and why you’re doing it. As we’ve discussed, there are many types of attribution, and not all of them are suitable for every team’s needs (i.e., requirements, but sometimes people are afraid of that word).
Sometimes, the who of attribution matters a lot. For a government seeking to use instruments of power (diplomatic, informational, military, economic), the who behind the keyboard is important. For example, the U.S. government frequently issues indictments against cyber adversaries (something I’ve taken an interest in) as well as levies economic sanctions against them. In those cases, attributing the who certainly matters.
Other times, the who of attribution doesn’t matter as much, and focusing on the how is sufficient. This isn’t always easy to discern, especially when geopolitical tensions are on the rise and we have fear, uncertainty, and doubt about who might target us. Situations like this provide us a good opportunity to reassess which threats we care about, make sure we’ve accounted for the corresponding TTPs, and reflect upon whether the who matters. If adversaries are on your network, does it really matter who is behind the keyboard if your main goal is just to get them out?
For example, as a company focused on defense and detection, we at Red Canary care much more about the how of activity so we can track adversary TTPs and better protect our customers. Of course, there are circumstances when the who matters to defenders, too… take red teams as an example. If I’m a defender, do I want to be spending my time responding to a red team if there are real adversaries on my network? No way! So, for many defenders, it may be important to track the who when it comes to red teams.
Each team is different in what it needs, and, as a result, everyone’s needs for attribution differ too. I encourage analysts to carefully consider their own team’s needs and then consider the different ways to do attribution to help them decide what makes sense.
Continue the discussion
With so many different approaches and needs for it, attribution is complex. If you’re interested in discussing this further and you’ll be at RSA, don’t miss my cooperative learning session on Thursday afternoon. If you’d like to chat with me or anyone else on the Red Canary team, feel free to reach out. See you in San Francisco!