The arms race between cybercriminals and cybersecurity professionals continues to escalate. And anyone watching the trajectory of this perpetual game of one-upmanship can see that this is a race towards implementing AI in the service of each side’s goals. For instance, a report by Nokia revealed that AI-powered botnets look for vulnerabilities in Android devices, then load data-stealing malware that is only detected after the damage has been done.

At the core of this competition is simple economics

Networks now incorporate multi-cloud environments that are dynamic and often temporary, SD-WAN connections to branch offices to support critical business applications, and an increasingly mobile workforce. At the same time, devices are proliferating inside networks at an unprecedented pace, from a multitude of different end-user devices to IoT technology. And consumer demand is accelerating the need for more powerful and responsive applications, which are in turn forcing workflows and transactions to span multiple data centers and ecosystems.

DX challenges for businesses

Digital transformation (DX) has completely upended years of security strategy. And due to issues like the cybersecurity skills gap, organizations can no longer afford to expand their security infrastructure organically.

Deploying, configuring, managing and operating security systems across multiple cloud environments, for example, can quickly overwhelm limited security resources. Traditional models need to be replaced with integrated systems that require fewer eyes and hands. Interoperability ensures that visibility and control extend across all security devices and span all network environments to close gaps, correlate threat intelligence and coordinate a response. And because of the speed and efficiency of cyberattacks today, this is only possible by implementing highly automated security solutions enhanced with machine learning to see and stop a threat before it can accomplish its goals.

DX opportunities for cybercriminals

Maintaining ROI in a cybercriminal enterprise requires lowering the overheard caused by constant innovation while increasing the efficiency and effectiveness of tools that penetrate defense systems and evade detection. Like their business counterparts, cybercriminals are increasingly turning to automation and machine learning to accomplish their goals.

DX efforts keep expanding the potential attack surface, providing new opportunities for exploits. To increase the efficiency of their “launch and pray” malware strategies, however, cybercriminals are increasing their odds by launching malware that runs on multiple operating systems and in multiple environments, and that can deliver a variety of exploits and payloads. By leveraging automation and machine learning, malware can quickly determine which payloads will be most successful without having to expose itself to detection through constant communications back to its C2 server.

AI is next

The goal for both sides of this battle is the eventual deployment of a solution that can adapt to unexpected environments and make effective autonomous decisions. This will require some sort of artificial intelligence. AI will allow businesses to deploy a self-defending security solution that can detect threats, close gaps, reconfigure devices and respond to threats without human intervention.

AI will also enable cybercriminals to deploy self-learning attacks that can quickly assess vulnerabilities, adapt malware to those weaknesses and actively counter security efforts to stop them. When combined with emerging threats like swarmbots, AI will be able to break down an attack into functional elements, assign them to different members of a swarm and use interactive communications across the swarm to accelerate the rate at which a breach can occur.

Shopping for an AI solution

Fortunately, the process of developing AI is very time- and resource-intensive, so few cybercriminals have been able to deploy even basic AI. But as AI becomes increasingly commoditized, it is a short hop to cybercriminal adoption.

The only defense against attacks enhanced with automation and machine learning is to have deployed those same strategies. And when AI becomes part of the malware toolkit, the organizations that fare the best will be those that have already begun to integrate AI into their defenses.

The security community has not adequately defined what constitutes an AI, which leaves buyers vulnerable. Many cybersecurity vendors at this year’s RSA conference, for example, claim to have AI capabilities. But in reality, most fall short because their underlying infrastructure is too small, their learning models are incomplete or the learning process has not had enough time to develop a sophisticated algorithm for solving problems.

When looking at a solution that claims to have an integrated AI engine, here are some questions to ask:

How many years have been spent developing this AI? True machine learning models require years of careful training to ensure that algorithms are stable, decision-making trees are mature and unexpected results are reduced to zero.

How many nodes does it use to learn and make decisions? This is not something that can be built in someone’s lab. It requires millions upon millions of nodes and a continuous feed of massive amounts of data before something even remotely like an AI can be developed.

How rich is your data? The lack of high-quality data is AI’s biggest drawback. Large and broad data sets are needed for successful implementation of AI. When this rich data is present, organizations are able to integrate AI into a comprehensive security framewok that creates centralized visibility, true automation and real-time intelligence sharing.

Fighting fire with fire

To win the cybersecurity war—especially during disruptive times like DX, and with limited cybersecurity resources—your best talent must be focused on the most critical decisions your organization faces. Achieving this requires handing over lower-order decisions and processing to automated systems.

However, not all AI is the same. Risk-based decision-making engines that are intelligent enough to take humans out of the loop not only need to be able to execute the “OODA loop” (Observe, Orient, Decide and Act) for the vast majority of situations it encounters but also actually suggest courses of action when a problem is discovered rather than merely relying on pre-defined ones. Only then can you confidently free up your valuable cybersecurity experts so they can concentrate on the more difficult decisions where human cognition and intervention is most required.

By Derek Manky, Chief, Security Insights & Global Threat Alliances at Fortinet – Office of CISO