CISOs are increasingly grappling with how organizations govern systems that can autonomously access data, make decisions, and trigger actions across enterprise environments. Security frameworks were traditionally built for rule-based software — not for systems that take independent actions, as agents do. Healthcare is becoming an especially useful test case because AI agents increasingly interact with highly sensitive data, regulated workflows, and mission-critical operations.

Data collaboration platform, Datavant, recently announced its membership in the AIUC-1 Consortium, a group developing standards for the safety, security, and reliability of agentic AI. Healthcare Innovation chatted with Datavant’s CISO, Dan Walsh, about the new cybersecurity realities in the agentic AI world.

Could you tell me about your organization?

We are a healthcare tech company. Our mission centers around what we refer to as longitudinal health record, providing the healthcare data pipelines for the medical records across the healthcare continuum. We’re embedded in a good portion of the US healthcare system, I think 70 percent of the US health systems.

Could you talk a bit about the AIUC-1 Consortium?

The AIUC-1 is an organization whose mission is to create a security compliance framework for agentic AI.  The traditional security measures and compliance frameworks that we have are falling a bit short. The AIUC-1 brings together industry leaders across various industries and ecosystems to put together a framework that is open, but also that will work for all as this field continues to evolve.

Could you discuss the new cybersecurity challenges healthcare organizations are facing now?

If we think about how current controls within the healthcare space are designed, there are new questions around whether an agent should have the same ability to do something as a human. How do we make sure that we manage them? It's different from an automation, which runs and does the same thing deterministically over and over again. These agents are now, in some cases, very non-deterministic.

How should organizations manage permissions for agents? What actions can they take, and not take, particularly with sensitive data in the environment? How do you continuously evaluate that the agents are doing the right things, and that they're performing the work as intended? And then, how do you explain sort of the path that they took? We might prompt an agent to do something and see the outcome, but how do we trace the actual action they took? What steps did they take? If they program something, what language do they use? Does that language have vulnerabilities? Can we have explainability when we have complex workflows and decisions within them, as we do in healthcare?

When you begin to introduce multiple agents, data sources, and tools — that's when the complexity really grows — and we just need to make sure we have traceability, transparency, and accountability, so we can govern in a very safe manner.

What are the steps to be taken to be in compliance?

Certainly, we want to start with HIPAA. HIPAA remains the critical foundation because it established requirements around privacy, access control, security, and auditability for protected health information. But it’s not enough. Because HIPAA was written before modern AI systems existed, and while some of the principles still apply, most organizations are increasingly needing more governance mechanisms around how we evaluate, monitor, and hold these systems accountable, and also how we measure the performance. The real question is, then, how do we operationalize that with these new AI workflow realities?

What challenges around AI do you feel concerned about?

I think it's just like with everything, with every new technology, the governance always sort of lags behind the technology. We've seen that in every single technological inflection point in history, so I think for us, joining the AIUC-1, we want to make sure that we are on the cutting edge.

Trust is one of our core principles at Datavant. It's embedded into our platform. It's non-negotiable. We also want to be able to take advantage of it, and we feel like we have a very unique position, given what we do in the healthcare industry and the talent that we bring. It's important for us to stay on top of this merging space as leaders and do it in a trusted, safe, and thoughtful way.

What do you feel healthcare organizations should be doing right now?

The basics still matter because AI is going to just speed things up. If you don't have MFA in place, if you don't have basic security controls in place…AI is going to make that worse. You have these models that can crawl and kind of ascertain where these weaknesses are.

Number one, you have to do the security basics. Number two, we need to make sure that humans always have clear accountability and oversight of this technology, whichever way an organization may take their AI journey. We also need to make sure that there's an inventory of what's running in your environment. I think a lot of times people are rushing to roll out this agentic capabilities, or even existing tools that are now introducing AI capabilities within them, but they don't have a really good inventory. If you don't understand what you're operating with, or what your company is using, you don't have a chance to start to begin to manage it. Once you have an inventory of it, I think it's important to understand its capabilities and the data it has access to. Based upon the data…what controls do we need? What could be done potentially with the data? Once you understand that, put together some of the other security capabilities—incident response, backups, patching...

It's still all the basics, but now it's just gone slightly faster, and because of the non-deterministic nature of this technology, there are a few additional steps that folks need to take.

Looking ahead, what kind of challenges or developments do you foresee?

We saw this with the cloud revolution 20 years ago. Amazon launched Amazon Web Services, and by 2013-2015, we began seeing large cloud breaches because people had misconfigured and they weren't patching the cloud. I think we're going to see those types of security incidents go faster this time. You're going to see a period of time where vulnerabilities are exploited because they're easy to chain together, and companies are going to struggle to keep up with patching. Companies have to shift the way that they think about how to operationalize some of these security practices.

I think it's going to resurrect potentially privacy conversations. Europe has done a good job with GDPR; we have nothing similar to that in the US.  Potentially, that's a conversation that may be had when people realize just the implications that AI has for privacy.

I think we'll probably see some more government action as these types of security incidents begin to unfold and begin to impact not only the US, but also other countries around the world.

Is there a place for governance in the government?

Whether it's the government, whether it is Health-ISAC, whether it is organizations like Datavant…we need to demonstrate that systems can perform reliably, consistently, and safely over time. Ultimately, trust has to be paramount.

Is there anything else that you think is important for our readership to know?

What we're hearing from our customers is that they want to have trust in the data. The other thing that we're hearing is that they want to know how performance is measured, evaluated, monitored, and improved over time. Governance accountability is very important. They are less interested in AI as the new cool kid on the block and more interested in how it will improve patient outcomes. How are we going to get better, high-quality information to the people who need it most? We're excited to be on this journey with the rest of the industry.