What is artificial adversarial intelligence?

artificial adversarial intelligence (noun)

A cyber defense approach that uses AI to replicate the behavior and decision-making patterns of cyber attackers and develop network defenses against them.

Advanced persistent threats — the most difficult types of cyberattack to detect — are carried out by high-level, often state-sponsored attackers using the most sophisticated tools and methods available. Now researchers at MIT are using artificial intelligence to learn how to thwart them.

They’re doing so through artificial adversarial intelligence — using AI to model the behavior of expert attackers, including how they learn and adapt in response to a target’s cyber defenses, Una-May O’Reilly, principal research scientist at the MIT Computer Science and Artificial Intelligence Laboratory, said in an interview with MIT News.

O’Reilly and her team are developing intelligent adversarial agents that have the computer skills of advanced human attackers. The researchers use those agents to attack networks in a test environment to uncover vulnerabilities in the cyber defense systems. The research team then uses AI and machine learning to analyze the AI agents’ activities, which allows them to develop measures against emerging attack methods and identify attackers’ likely countermeasures to new cyber defenses.

O’Reilly likened the scenario to the cat-and-mouse game played out in old Tom and Jerry cartoons, where the two adversaries continually learn from each other. “We work to replicate cyber versions of these arms races,” O’Reilly said

3 Questions: Modeling adversarial intelligence to exploit AI's security vulnerabilities