The Cisco Stay Community Operations Middle (NOC) deployed Cisco Umbrella for Area Title Service (DNS) queries and safety. The Safety Operations Middle (SOC) group built-in the DNS logs into Splunk Enterprise Safety and Cisco XDR.
To guard the Cisco Stay attendees on the community, the default Safety profile was enabled, to dam queries to recognized malware, command and management, phishing, DNS tunneling and cryptomining domains. There are events when an individual must go to a blocked area, such a stay demonstration or coaching session.
Through the Cisco Stay San Diego 2025 convention, and different conferences we’ve labored up to now, we’ve noticed domains which can be two to a few phrases in a random order like “alphabladeconnect(.)com” for example. These domains are linked to a phishing marketing campaign and are generally not but recognized as malicious.
Ivan Berlinson, our lead integration engineer, created XDR automation workflows with Splunk to determine High Domains seen within the final six and 24 hours from the Umbrella DNS logs, as this can be utilized to alert to an an infection or marketing campaign. We seen that domains that adopted the three random names sample began to exhibiting up, like 23 queries to shotgunchancecruel(.)com in 24 hours.
This bought me considering, “Might we catch these domains utilizing code and with our push to make use of AI, may we leverage AI to seek out them for us?”
The reply is, “Sure”, however with caveats and a few tuning. To make this attainable, I first wanted to determine the classes of information I wished. Earlier than the domains get marked as malicious, they’re often categorized as procuring, ads, commerceor uncategorized.
I began off operating a small LLM on my Mac and chatting with it to find out if the performance I would like is there. I advised it the necessities of needing to be two-three random phrases, and to inform me if it thinks it’s a phishing area. I gave it a number of domains that we already knew have been malicious, and it was in a position to inform that they have been phishing based on my standards. That was all I wanted to start out coding.
I made a script to tug down the allowed domains from Umbrella, create a de-duped set of the domains after which ship it to the LLM to course of them with an preliminary immediate being what I advised it earlier. This didn’t work out too effectively for me, because it was a smaller mannequin. I overwhelmed it with the quantity of information and rapidly broke it. It began returning solutions that didn’t make sense and totally different languages.
I rapidly modified the conduct of how I despatched the domains over. I began off sending domains in chunks of 10 at a time, then bought as much as 50 at a time since that gave the impression to be the max earlier than I assumed it will develop into unreliable in its conduct.
Throughout this course of I seen variations in its responses to the info. It is because I used to be giving it the preliminary immediate I created each time I despatched a brand new chunk of domains, and it will interpret that immediate in another way every time. This led me to switch the mannequin’s modelfile. This file is used as the foundation of how the mannequin will behave. It may be modified to alter how a mannequin will reply, analyze information, and be constructed. I began modifying this file from being a common function, useful assistant, to being a SOC assistant, with consideration to element and responding solely in JSON.
This was nice, as a result of now it was constantly responding to how I wished it to, however there have been many false positives. I used to be getting a couple of 15–20% false constructive (FP) price. This was not acceptable to me, as I wish to have excessive constancy alerts and fewer analysis when an alert is available in.
Right here is an instance of the FP price for 50 at this level and it was oftentimes a lot increased:
I began tuning the modelfile to inform the mannequin to provide me a confidence rating as effectively. Now I used to be in a position to see how assured it was in its dedication. I used to be getting a ton of 100% on domains for AWS, CDNs, and the like. Tuning the modelfile ought to repair that although. I up to date the modelfile to be extra particular in its evaluation. I added that there shouldn’t be any delimiters, like a dot or sprint between the phrases. And I gave it detrimental and constructive samples it may use as examples when analyzing the domains fed to it.
This labored wonders. We went from a 15–20% FP price to about 10%. 10% is significantly better than earlier than, however that’s nonetheless 100 domains out of 1000 that may have to test. I attempted modifying the modelfile extra to see if I may get the FP price down, however with no success. I swapped to a more recent mannequin and was in a position to drop the FP price to 7%. This reveals that the mannequin you begin with is not going to at all times be the mannequin you find yourself with or will fit your wants essentially the most.
At this level, I used to be pretty pleased with it however ideally want to get the FP price down even additional. However with the mannequin’s present capabilities, it was in a position to efficiently determine phishing domains that weren’t marked as malicious, and we added them to our block record. Later, they have been up to date in Umbrella to be malicious.
This was an important feat for me, however I wanted to go additional. I labored with Christian Clasenour resident Umbrella/Safe Entry professional and was in a position to get a slew of domains related to the phishing marketing campaign and I curated a coaching set to wonderful tune a mannequin.
This process proved to be tougher than I assumed, and I used to be not in a position to wonderful tune a mannequin earlier than the occasion ended. However that analysis remains to be ongoing in preparation for Black Hat USA 2025.
We’d love to listen to what you assume! Ask a query and keep linked with Cisco Safety on social media.
Cisco Safety Social Media
Share:
