AI is creating a security problem that most companies don’t have the staff to deal with, says an AI researcher

Companies may have cybersecurity teams in place, but many are still unprepared for the actual failure of AI systems, says an AI security researcher.

Sander Schulhoff, who wrote one of the first guides to timely technology focusing on vulnerabilities in AI systems, said on an episode of “Lenny’s Podcast” published Sunday that many companies lack the talent needed to understand and remediate AI security risks.

Traditional cybersecurity teams are trained to troubleshoot and fix known vulnerabilities, but AI doesn’t behave that way.

“You can fix a bug, but you can’t fix a brain,” Schulhoff said, describing what he believes is a mismatch between the way security teams think and the failures of large language models.

“There is a disconnect between how AI works and traditional cybersecurity,” he added.