Where’s your PIV card, HAL? What happens when AI gets an employee ID

What happens when an AI system stops being something a federal employee uses and starts doing things on that employee’s behalf?

That’s the question behind the latest episode of Fed Thread, where Ryan Galluzzo and Bill Fisher from the National Institute of Standards and Technology explain what happens if an AI agent is going to operate inside a federal environment.

How do you know who it is, what it’s allowed to do and who’s responsible for what it does?

Here’s a snapshot of the conversation:

1. Should your AI agent borrow your credentials?

It’s a tempting shortcut to let your AI agent use the same credentials as you. But Fisher says that’s exactly the kind of workaround organizations need to avoid, the same way you wouldn’t share your login and password with a colleague.

Unfortunately, it takes more than an absolutist stance on the matter. If the secure way to use an AI agent is too difficult, Fisher says, people may find another way around it and still misuse the credentialing.

The solution, he argues, is to make the secure path the easy one, with agents getting their own identities and credentials rather than borrowing ours.

1. How much authority should your AI agent have?

Just because your AI assistant works for you, that doesn’t mean it gets to be you.

Galluzzo takes the familiar cybersecurity principle of least privilege and applies it to AI agents, which means an agent shouldn’t automatically inherit everything its human sponsor can access. Rather, its permissions should be narrowed to only what it needs for a specific task.

1. Why does an AI agent need an identity at all?

It’s not just about controlling access. An identifiable agent also creates a trail:

What did it do? When did it do it? What was the result?

Perhaps most importantly, was that action taken by the human or by the agent acting on the human’s behalf?

Without that distinction, accountability gets murky, Galluzzo and Fisher explain. With clear agent identities, however, agencies have a better chance of reconstructing what happened when something goes wrong.

1. Can the AI agent be fooled?

Fisher offers a memorable example: An AI agent is asked to sort through hundreds or thousands of USAJOBS applications, but one applicant hides instructions in white text inside a résumé, telling the agent to put that résumé at the top.

The human reviewer can’t see the hidden text that AI can, and so the unqualified résumé leapfrogs all the qualified ones, and the humans in charge are left wondering how it made it to the top of the pile.

Because all incoming instructions are treated equally by AI, an agent can produce results that have been manipulated to skirt its intended filters.

1. Who is running the AI?

This is where technology meets governance. Employees using AI tools outside of an organization’s approved processes can potentially send information into systems the organization hasn’t vetted.

The answer isn’t simply to tell employees not to use AI. Galluzzo and Fisher instead say that agencies need clear policies, approved use cases and a way to understand the risks.

1. What happens when AI acts faster than humans can verify the action?

“Trust but verify” sounds reasonable until an AI agent can take actions in seconds that could take a human days or longer to review.

At some point, there may be decisions where verifying every action isn’t practical. That means agencies will have to think carefully about where human judgment remains essential, how much risk is acceptable and which actions should never be left entirely to an agent, says Galluzzo.

Ultimately, Fisher adds, if something goes wrong, you don’t get to blame your AI. The human (and the organization itself) still owns the decision.

Want to hear the finer details of the conversation? Listen to the full Fed Thread episode.

And for extra credit reading, check out some of NIST’s latest resources:

1. NIST concept paper: “ACCELERATING THE ADOPTION OF SOFTWARE AND AI AGENT IDENTITY AND AUTHORIZATION”
2. NIST summary of comments on the concept paper
3. Galluzzo and Fisher blog post on the need for agentic AI identities

Catch more conversations on podcast platforms, YouTube or on the Fed Thread hub page.

Copyright © 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.