Ep. 279 Avoiding Agentic AI Pitfalls in Federal Digital Transformation

A recent study from Carnegie Mellon University is titled "AI Agents Fail at Office Tasks Nearly 70% of the Time."

Federal agencies are adopting Agentic AI for the efficiency it can deliver. Unfortunately, many do not realize that Agentic AI is prone to operational risks, ranging from technical glitches to legal complications to accidental database deletion.

When Agentic AI causes problems at a federal agency, there can be lives at stake.

Today, we sat down with Travis Rosiek, Rubrik's Public Sector Chief Technology Officer. During the interview, he explores the federal challenges of implementing Agentic AI, building an Agentic AI inventory, and making Agentic AI visible, auditable, and reversible.

CHALLENGES

Everyone —from a systems administrator to an agency administrator —knows that data must be backed up. However, very few understand that Agentic AI is a collection of agents that can be attacked, just like a database.

Rubrik offers the capability to reassure users that Agentic AI can be reversed if malicious actors enter the picture.

STARTING POINT

Most cybersecurity professionals agree that one starts by understanding a system's apps, data, and connections. Five years ago, it was easy; getting a grasp on what Agentic AI connects to is a much more intangible concept.

During the interview, Travis Rosiek unpacks Rubrik's history and its unique ability to understand complex systems.

CAREFUL

In a rather shocking statement, Rosiek says one should approach introducing technology with the assumption that it will fail. This is not a pessimistic approach, but a nuanced understanding of how complexities in current systems can lead to unintended consequences.

Rosiek advises starting with the end goal in mind, planning for worst-case scenarios, and building trustworthy AI architectures to mitigate risks and ensure reliable operations.