Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

How is AI transforming cybersecurity?

AI is starting to feature a lot in cybersecurity. Here are a few areas where AI can help provide value.

[Images: Adobe Stock / PRIM]

Profile photo of Darren Anstee

Chief Technology Officer for Security · NETSCOUT

The first question is: “Why am I spending time writing this?” I typed the title of this article into a well-known generative AI platform, and it spat out an 800-word article in about five seconds. Although the result was a fine read and it didn’t say anything wrong, it was also quite vague in what it did say.

Why do I think this is interesting? Mainly because I see a lot of unclear messaging and marketing around AI that enthuses about the benefits it will bring, without really qualifying anything. Seeing the similarities, part of me wonders if we are already caught in a cycle of AI-generated content based on AI-generated content—who knows?

Moving on from that—and hopefully being a bit more specific—AI is starting to feature a lot in cybersecurity and in the conversations I am having with our customers. These customers have specific needs and want to know where AI can provide value. There are a few areas where AI can help a lot. Here are a few:

CYBER THREAT INTELLIGENCE

Firstly, let’s look at cyber threat intelligence (CTI). CTI has been around forever, and we all use it in multiple aspects of our security stack. In fact, in many cases, the effectiveness of our defensive technologies is determined by the quality of the CTI we are using. This is where AI plays an increasing role.

Some threat intelligence providers are now using AI to sift through the vast amounts of data they see to deliver higher-quality threat intelligence at a faster speed. Given how rapidly threats develop and how widely organizations are targeted, timelines in threat intelligence are key, as is its efficacy (low false-positive and false-negative rates).

By using better CTI, our existing security products become more effective, reducing operational costs and business risks—a win for AI—and businesses generally.

INCIDENT RESPONSE

Secondly, when it comes to incident response, AI is showing a lot of value. There are variations in how AI is being used across different platforms, but there are three main areas. Some security information and event management(SIEM) and security orchestration, automation, and response (SOAR) systems are using AI to collate, enrich, and prioritize information so that operators have an easier-to-assess, pre-built package of data to consult for any event. This allows them to process more events, more consistently, helping to ensure better overall triage, investigation, and containment processes.

Alongside this, some systems also use AI to identify events that can be dealt with programmatically, reducing the workload of operators by dealing with them automatically. And, more recently, some platforms have introduced generative AI (chatbot) assistants that can help operators both analyze events and determine the best path forward.

Again, in all of these, we see AI being focused on decreasing operational costs and business risks—another win-win.

DETECTION OF THREATS

A third area where AI is making a difference is in the detection of threats (anomalies). Here, there are a wide variety of techniques being used across packet data, session metadata, log data, and even event data. Some vendors are implementing expert systems that can classify a threat and recommend or apply the best ways of containing or mitigating it. This reduces the need to use expert resources, allowing them to focus elsewhere, as well as accelerating response—this can be a very effective use of AI.

Some organizations use more general AI to identify anomalous behaviors, but there are a couple of challenges, the first and most important of which is data quality. Some of you may remember Garbage In, Garbage Out (GIGO) from 1980s computer science (showing my age) to describe the poor outcome you will get if you feed an algorithm poor data. This is still true of AI today, regardless of how sophisticated the implementation is. Feeding in poor data can lead to the output being less likely to be right. This happens, more often than not, when the data being used to drive the AI algorithm is more aggregated or abstracted from what ‘actually’ happened.

This brings me to the second challenge when using AI for anomaly detection: explainability. For example, when our security systems detect an event based on an indicators of compromise (IoC) match, we know categorically why the event was raised. We also know the nature of the threat based on the contextual information we have for the host generating the violation and the IoC. With some AI, anomalies can be raised that are not explained—we are told ‘what’ is anomalous but not ‘why.’ This is a problem for operators as they don’t know how to move forward without doing a lot more digging, which takes a lot more time.

Sometimes this digging may lead to finding a threat that would otherwise have gotten by—which is a big win—but in other cases, it doesn’t, and there’s no way to tell at the start.

Every time we find a threat that would have been missed, trust is built in AI; every time we find out we’ve wasted time, trust is eroded. We must keep the balance on the former by using the best data and the best algorithms to get the best from AI. When it comes to anomaly detection, AI can deliver a lot of value.

There’s no doubt that we will continue to see AI being used more in cybersecurity in the areas above, and others. AI will enhance our capabilities, but let’s not forget that it’s not a magic bullet. AI can certainly make us more efficient and can help us find things that would otherwise have been missed—but only when used in the right way.


Darren Anstee is CTO for Security at NETSCOUT, which helps assure digital business services against disruptions.