As with any IT system, generative and agentic AI implementations in business settings need to be secure.
On the latest episode of the Targeting AI podcast from AI Business, Oren Michels, founder and CEO of AI data and access management startup Barndoor.ai, discusses the cybersecurity implications of the rapidly proliferating array of new enterprise AI systems.
The podcast episode, hosted by Esther Shittu, was recorded live at the AI Summit in New York City, where Michels was scheduled to appear on a panel to discuss enterprise-grade AI security.
On the podcast, Michels said enterprises need to take a security approach to agentic systems that differs from traditional IT cybersecurity strategies. The old paradigm focused on preventing outsiders from infiltrating the system and causing harm.
Now, with AI agents, people from inside the organization are using agents to pursue and complete tasks they’re supposed to do.
“But the AI [agents] might or might not do the things that are expected for those outcomes to happen,” Michels said. “So, we want to make sure that security is provided where the agency is based, not just on the identity of the human, but also takes into account aspects of the agent itself.”