These are the droids you’re looking for: Tireless, uncomplaining, able to work 24 hours every day, responding instantly to threats against the Empire… or rather, threats against an enterprise, service provider, or telco.
We’re talking about artificial intelligence software, equipped with machine-learning algorithms, designed to work as part of a comprehensive security suite for the Security Operations Center (SOC).
“AI helps us go from hindsight to foresight,” says Vinod Peris, Senior Vice President at industry giant CA Technologies. “If you take things like malware detection, they used to be pattern-based. You were matching some patterns. Now, you’re trying to actually look at behaviors: behaviors of applications, behaviors of end-points.”
From matching patterns to looking at behavior to anomalies. “When you look at the SOC, there are certain areas where AI can help,” adds Slavik Markovich, CEO of security innovator Demisto.
“First of all identifying the needle from the haystack. That’s one of the easiest things you would do with AI, because AI is used a lot to take a lot of data and cluster it, and classify it, and try to find the outliers.”
Another application of AI, says Markovich, is to reduce the repetitive workload for security analysts. “Once an analyst identifies an alert, there are a lot of mundane tasks where you basically do the same thing over and over again, like get the reputation of a certain IP, check the provenance of a file, check logs, and all of those things.”
AI can do those tasks instead, in an adaptable day: “This is something AI is really good at, by looking at already-existing actions in the history and learning from them, and then feeding it back to the analyst whenever it’s needed.”
Greg Fitzgerald, CMO of security startup JASK, agrees about AI helping SOC analysts, and adds that algorithms can augment the decision-making process.
“For example, AI can correlate disparate alerts and events that may be happening in different locations or at different severity levels that a human may not even pay attention to,” he says. “So, AI captures all that, creates the linkages between those different alerts, and is able to aggregate those into things we actually call an insight.
“An insight is not the resolution that this is definitively happening,” explains Fitzgerald. “It’s a collection of the activity that now a human can use the subjective capability to determine ‘This is relevant for me and my organization at this time.’ ”
Key AI Technologies for the SOC
Researchers have developed dozens of AI technologies, algorithms, models, and learning techniques. Two of the most common used in the security field are machine learning and predictive analytics. Machine learning is a data-intensive technique that lets software progressively improve performance on specific tasks, particularly those involving classification of data, and predictions about future events. In the AI domain, predictive analytics is used to swiftly detect anomalies or new patterns in the data, and make recommendations based on those anomalies and patterns.
Demisto’s Markovich adds that security sometimes uses unguided deep learning, which is based on neural networks, as well as the guided techniques above. Deep learning, which mimics the way humans behave more than machine learning, can produce results similar to what a human SOC analyst would do. “You have deep learning, you have the guided learning; people are using all of that.”
At the end of the day, there’s a commonality: “Some software gets a lot of samples, and then based on various attributes classifies them,” explains Markovich. “Others look at lots and lots of events, and then try to correlate and find the right things to do. Companies like Demisto look at analysts’ intents and actions and derive the value from there.”
It’s all changing the world, says Sam Liang, CEO and Founder of AISense, a company focusing on using AI and speech recognition to create intelligent and contextually-aware mobile tools to enhance professional productivity. “You know, Marc Andreessen said software was eating the world, but now it’s actually that AI is eating the world.”
“All the major companies have got AI,” explains Liang, “AI is changing people’s lives in a lot of different perspectives in terms of, you know, speech recognition, image recognition, medical imaging. You can have AI to look at your MRI or x-ray to quickly detect a cancer.” Or to correlate data to detect a malware incursion or a hacker’s attack long before a human analyst would sense that something’s wrong, he adds.
Faster Time to Triage
JASK’s Fitzgerald agrees that AI is speeding up the detection of problematic events. “The biggest impact that AI is making in the SOC is faster time to triage, because in the end, that’s exactly what the human is trying to do. We see lots of reports that talk about, a compromise hasn’t been found for 100 days, or 150 days, or more.” AI is cutting that time-to-detections from months to minutes, he adds.
“It’s definitely in the detection,” comments Demisto’s Markovich. “When you look at detection, there’s a mass of events out there, and it’s almost impossible for a human to look at all of them. AI is doing a pretty decent job in highlighting what’s interesting.”
Still, Markovich points out, AI can’t do the job alone. “We’re not in a place where you can take the human out of the equation. AI can need to highlight the right stuff, but then allow the human to actually look at it, interact with it, and decide if it’s a real incident or a false positive.”
“AI will help in reducing the time to triage,” says CA’s Peris. “If you have a SOC issue, the first thing people do is, they get on a call and they try to figure out what’s the root cause and go through the analysis. With AI, you could have this at your fingertips, hopefully even before that issue happens.”
Peris adds that AI will play multiple roles in protecting networks, applications – and people. “Facebook recently had to put in place a system that could look at, and prevent, fake news. They hired 10,000 people for this. This is obviously just a quick reaction. What AI will eventually do is allow machines to do that first level of analysis, so it will cut down the number of people that you need.” And perhaps do the job more accurately, with a more consistent set of rules.
What’s Next for AI in Security?
CA’s Peris sees a bright future for end-point behavioral analytics. “If somebody steals your password, unless they know exactly your access patterns, your system can detect that it’s not you through behavioral analytics. Similarly, you can do the same thing for applications. If you knew the application’s behavior, and let’s say we as application developers gave you a signature of the application’s behavior, you could factor that into figuring out when an application is compromised. “
“More optimization,” says Demisto’s Markovich. “You’ll see a lot of faster algorithms, maybe more accuracy, fewer false positives. The big bet is actually in unsupervised learning, that is, deep learning. Throw the bunch of events on enough computing power to let AI learn by itself. That will eventually get to a place where AI can actually identify the real positives and not false positives.” This level of deep learning might be 20 years away, says Markovich, but “that would be the real game-changer.”
Fitzgerald from JASK believes that “AI for SOC will expand beyond just the analytics of the alerts, and the logs, and the information that’s being ingested” and head into the ability to respond without human supervision. “To date, most things are still provided to what’s called a playbook, or something that allows a human to start taking physical action.”
“In the next couple of years, the SOC analyst will start to trust the decisions that are being made,” adds Fitzgerald, “and allow AI to automatically make the configurations that rectify the situation without the analyst’s involvement, but with the analyst’s supervision.”
In other words, soon we’ll let AI determine if those really are the droids you’re looking for, in an Empire far, far away.