What Happens When Agentic AIs Talk to Each Other?

Agentic AI is no longer a futuristic thought experiment. It’s here, operating across internal systems, automating workflows, and making decisions in milliseconds. But in this new reality, one big question remains: what happens when autonomous agents start interacting with each other?

That’s the question we posed earlier this year in our blog 5 Things Security Leaders Need to Know About Agentic AI. The response sparked thoughtful commentary and engagement across platforms (especially LinkedIn) and helped us realize it was time to open the conversation even further.

Experts on Experts: Commanding Perspectives is a new short-form video series that pairs Rapid7’s internal leaders in Q&A-style interviews – making space for honest expert conversations on the issues security leaders are actively wrestling with. With that, we're pleased to share that our first episode is now live!

In this 10-minute conversation, Craig Adams, SVP, Chief Product Officer, sits down with Laura Ellis, VP of Data & AI at Rapid7. The topic is Agentic AI – specifically the challenges of autonomous agents that interact, iterate, and potentially act beyond their intended scope.

Some discussion highlights:

• Why containment zones and “kill switches” might not be enough

• How and where indirect prompt injection is already showing up inside integrated systems

• Why governance must go beyond observability, and what that actually means

• Laura's thoughts on “vibe coding” and the risk of well-meaning business users creating security blind spots

Tune into our first episode:

 
...And stay tuned! The next episode will feature Jon Hencinski – VP, Detection & Response – on MDR, risk prioritization, and how D&R is evolving in hybrid environments.