Identity After Dark: What 30 Practitioners Said When No One Was Pitching

On June 17, we hosted a live recording of the Identity at the Center podcast at Identiverse. Jim McDonald and Jeff Steadman ran the room. I was there as a panelist. About thirty senior IAM practitioners showed up on a Wednesday evening: directors, architects, CISOs, consultants. They skipped whatever else was happening on the Strip. They spent the next hour and a half being genuinely honest with each other.

No slides. No agenda I could hide behind. No prepared answers that let you sound smarter than the question deserves.

Here's what I heard.

AI identities are not the problem. The inventory is.

The first question of the night was about panic. Should IAM teams panic about AI identities? Honest answer: a little. Not because AI identities are inherently unmanageable, but because most organizations don't actually know what they have. Jeff said it clearly: "Whatever number you pick is already out of date." And then he said the thing that matters: start with an inventory.

That's it. That's step one. Not a new platform, not a new standard. Just a clear view of what you're running and what it has access to. We've been saying this about human identities for two decades. We're saying it again about AI agents. At some point you stop being surprised that the answer is the same.

I do think AI will help us manage this at scale. The future of identity security will probably be secured by agents more than by humans reviewing entitlements in spreadsheets. But we are not there yet. Right now, visibility comes first.

Access reviews are not dead. They're just a sign of something.

The room had a lot to say about access reviews. Someone put it plainly: the number of access reviews your organization runs is a sign of immaturity. The more blanket certifications you need, the further you are from a policy-driven, usage-based governance model.

I think that's right. We've seen this internally. A very small percentage of entitlements are ever actually revoked in a standard access review campaign. People approve what they don't understand because the friction to re-request is too high if they get it wrong.

The idea that cut through the room was simple. What if you revoke the access immediately, but let the user get it back without an approval workflow? Build the trust in the re-grant process, and suddenly the revocation stops being the scary part.

That's not a vendor pitch. That's a design principle. And it came from the room, not from me.

Vendor relationships: the conversation nobody wants to have

Jeff made an argument I want to repeat. He said we need to normalize it being okay to rip and replace a vendor. The industry treats bad vendor decisions like career-ending events. They are not. They are just decisions that age out.

He's right. And I'll add something from the vendor side: some of our fastest implementations have been with customers who are replacing an existing product. They've already done the hard work. The groundwork is there. Standards make the second pass cleaner.

I'm not in the business of encouraging customers to replace products for the sake of it. But if something is not delivering, replacing it is a sign of mature decision-making, not failure.

Agentic authentication: delegation, not impersonation

The room landed on a framing I found genuinely useful. When we're talking about AI agents and what they can access, the model should be delegation, not impersonation. An agent gets a specific set of permissions to do a specific task. It doesn't inherit your credentials. It doesn't act as you.

The workload identity analogy holds up well here. Think about how ephemeral credentials work in a DevOps pipeline. They last for the duration of the job, then they're gone. That's where agent identity is going: not service accounts that sit open for years, not human-speed access reviews, but just-in-time credentials scoped to the task.

We don't need a new standard for this. We need to use the standards we have, implemented correctly, with the right logging and traceability. Jim put it well: delegation, not impersonation. The room was the point.

The episode is now live on the IDAC feed. You can listen to the full conversation, including the questions we didn't answer, and there were a lot of those.

I wrote up the unanswered questions separately. The room submitted more than any one episode could cover. I worked through my responses to the ones that didn't make it. You can get them here.

What I took from that evening wasn't a new argument or a better slide for a deck. It was a reminder that the most useful conversations in this industry happen when nobody is pitching. The room in Las Vegas did that. I hope the episode does the same for whoever listens.

*Bart Allan is the General Manager at Bravura Security. Bravura Security hosted the live recording of IDAC episode 442 at Identiverse 2026.*