The Internet has enhanced communications, increased commerce, and brought people together socially. Unfortunately, it has also enabled malicious activity with data breaches, ransomware, destroyed systems, and the Dark Web. Cyberattacks have become so common that only the large ones make the news now. The United States is arguably the most "wired" country in the world, with everything from cars to refrigerators to security cameras connected online, making us also the most vulnerable. Because the open Internet is driven by cost and speed and not by security, continual cyberattacks have pushed us into a new kind of Cold War — with artificial intelligence (AI) serving as the basis of this arms race.
From Moonlight Maze in the late 1990s to the recent SolarWinds attack, we have seen malware and ransomware planted in our infrastructure and systems. Nation-states have staged cyberattacks, often as a prelude to military actions. Attacks launched from the open Internet are at a constant level of activity, just below armed conflict.
We think of cyberattacks in terms of router configurations or malware code, but the tremendous amounts of communications traffic make cybersecurity a field of data science. All the new sensors and Internet of Things devices produce tremendous amounts of data that can be analyzed to detect adversary activity. Such massive volumes of data need analytic techniques to synthesize the essence of the activity for human understanding and decision-making.
The use of AI to analyze these massive amounts of cyber data and capabilities is growing exponentially. In 2016, when the AI-driven Alpha Go beat the world Go champion, it was a "Sputnik moment" about the growth of AI. A year later, China released its New Generation Artificial Intelligence Development Plan to be the world leader in AI by 2030. In 2020 at a DARPA virtual dogfight, the AI pilot beat the human pilot. It is now clear that AI has progressed quickly to have real-world security implications.
Cyberattacks are now constant and range from annoying to devastating. There is still a significant lag between attack and detection — and we need to use AI to improve defenses and reduce that gap. Previously, rules-based systems were applied in cybersecurity to detect malware signatures and look for known insider threat patterns. But AI's self-learning techniques are now being used to look for unknown insider threat patterns and other malicious activity. Significantly, the methods can even learn based on data that may already contain the threat activity. In addition, AI can synthesize the difference between normal router outages in the open Internet and botnet attacks.
In this Cyber Cold War, we must always assume that our defenses have been breached and our adversaries are already in our systems. This assumption is the driver for zero trust — an essential methodology to defend ourselves in this new Cold War. All organizations must move from perimeter security to follow the principles of zero trust — to bake in role-based access controls for every resource and provide secure communications throughout. We must also not assume that the open Internet is benign "plumbing." The zero-trust concept must also be extended into the open Internet to "defend forward" beyond our network perimeters.
Finally, we need to extend our security monitoring to leverage AI to search for unknown patterns of malicious activity. AI is the arms race to analyze more data with greater speed and provide transparent identification and recommendations for human decision-makers.
Cyber resiliency must not be viewed as a costly optional add-on but must be considered an essential part of doing business. We must recognize that the more governments develop offensive techniques, the more vulnerable our infrastructure is on defense. Cyber will be a part of every future military action, and in the world of cyberattacks, there are no non-combatants.