Enable breadcrumbs token at /includes/pageheader.html.twig

Defending Against Adversarial AI Capabilities

Subject matter experts discuss concerns surrounding artificial intelligence and its use by bad actors.

Panelists at TechNet Cyber 2023 discuss the impact of artificial intelligence on cybersecurity. Credit: Michael Carpenter

Artificial intelligence (AI) remains one of the most popular topics of discussion inside and outside the military world. Though the benefits of the technology continue to be showcased, risks are still involved, explained a panel of speakers during a discussion on convergence and cyber synergy for success at TechNet Cyber 2023 held in Baltimore on Wednesday.

“I think we have a few steps to consider before we can do AI/ML [machine learning] at scale for cybersecurity,” said Wendell (Dell) E. Foster Jr., executive director at the Joint Force Headquarters – Department of Defense Information Networks (DODIN). He explained that although the department has advanced in censoring, aggregating and storing data, there is still much to be done before leveraging the highly talked-about modern solutions. “We believe that there will always be a human in the loop for a lot of those key decisions in cyberspace operations,” he added.

“AI could be used for good; it can also be used for evil,” noted David P. Luber, the National Security Agency’s deputy director for the Cybersecurity Directorate. “So, as we think about how AI can help you in analyzing malware, AI can also be used by a bad actor.” Luber expanded on the critical need to defend against adversarial AI capabilities, which add risk to national security systems.

Data is at the heart of the issue, explained David W. McKeown, senior information security officer, deputy chief information officer for cybersecurity and acting principal deputy chief information officer for the Department of Defense. While speaking on all the different cloud vendor environments, McKeown mentioned that the department cannot afford to duplicate data. “We’ve got to figure out a way to make that data available to all four of those clouds and searchable to all four of those clouds so that we can run the AI analytics,” he said.

Major vendors such as Google, Oracle, AWS and Microsoft will be brought to the table to build custom clouds for the department, McKeown said. “Our next step is to figure out what problems we want to turbocharge by applying AI to it,” he noted.

The big risk lies in bad actors using AI capabilities to hack systems and steal data. “The adversary having this is a big concern,” McKeown continued. “One time I was asked, am I more concerned about quantum or AI, and at the time I didn’t really know that much about AI, so I said quantum … but AI is scary in the capabilities that it has.”