Securing Agentic AI: A Video Commentary Series from Numberline Security
Agentic AI is moving fast, and so are the security questions it raises. As AI systems take on autonomous roles, make decisions, and interact with sensitive resources, security teams need to think carefully about how Zero Trust principles apply in this new landscape.
Numberline co-founders Jason Garbis and Jerry Chapman have put together a three-part series of short video commentaries to help cut through the noise. Each installment tackles a specific, practical question: no theory for its own sake, just the kind of clear-eyed analysis you’d expect from practitioners who work on this every day.
Part 1: The Overprivilege Problem, and What to Do About It
Overprivileged agents are one of the most common and most dangerous mistakes in agentic AI deployments. In this first video, Jason and Jerry dig into why least privilege matters more than ever for AI agents, what tends to go wrong, and how basic security controls turn out to be surprisingly effective. (Plus: why you really shouldn’t be like Jerry.)
Part 2: Are AI Agents a Different Kind of Identity?
It sounds like a philosophical question, but it has very real security implications. Are agentic AI systems fundamentally different from the human and machine identities we already manage, or just more of the same? Jason and Jerry debate this one with the help of an unusual analogy: the four states of matter.
Part 3: A Taxonomy of AI Agents, and the Security Risks of Each
Not all agents are alike. The security concerns for a simple task-automation agent are very different from those of a multi-agent orchestration system with broad access to enterprise resources. In the final installment, Jason and Jerry walk through a practical taxonomy of agent types and lay out the primary risks you need to account for in each case.
Whether you’re just starting to think about agentic AI security or already knee-deep in it, these videos are a quick, worthwhile investment of your time.