
The healthcare industry has spent the last few years talking about generative AI. People have debated use cases, explored copilots, tested document summarization, and explored productivity gains. The next conversation will be different.
We’re entering the era of AI agents.
Recent announcements from Epic around its Agent Factory platform and expanding portfolio of AI-powered assistants signal a significant shift. Instead of simply generating content or answering questions, AI agents are beginning to perform actions. Epic’s vision includes agents that can reason through tasks, access multiple systems, and execute actions under defined governance models.
This is exciting. It is also a security and governance inflection point.
The question is no longer whether AI can generate a response. The question is whether we are prepared for AI that can act.
From Copilots to Colleagues
Traditional generative AI operates largely as a recommendation engine. A user asks a question, receives a response, and ultimately remains responsible for the decision.
Agentic AI attempts to change that model.
An AI agent may:
- Review information across multiple applications
- Initiate workflows
- Schedule appointments
- Process prior authorizations
- Draft appeals
- Trigger notifications
- Update records based on predefined rules
Epic’s Agent Factory platform is designed to enable healthcare organizations to build and manage these types of agents within operational workflows.
On paper, the benefits are significant. Reduced administrative burden, faster patient service, improved clinician efficiency, and streamlined operational processes are all compelling outcomes. According to Epic, health systems using AI-enabled workflows through Epic have already reported measurable results, including reduced documentation time, fewer denials, and enhanced patient experiences.
However, every action an agent takes introduces new security considerations.
Security Models Built in the Past Were Designed for Humans
Most healthcare security programs were designed around a relatively straightforward premise: authenticate a user, authorize access, monitor activity, and respond to violations.
That model becomes more complicated when the “user” is an AI agent.
Consider a prior authorization agent that accesses an EHR, payer portal, document repository, email system, and scheduling application.
Questions a CIO and CISO should ask include:
- What permissions does the agent have?
- Who approved those permissions?
- Can the agent invoke another agent?
- How do we audit every action the agent performs?
- What happens if the agent makes an erroneous decision?
- How do we terminate agent activity instantly if risk is detected?
These are not future-state concerns. They are present-day governance questions.
The challenge is that AI agents often require broader access than traditional software automation because they must reason across systems and dynamically adapt to changing circumstances.
The more capable the agent becomes, the more important governance becomes.
Security Cannot Be an Afterthought in the Age of Agentic AI
As AI agents evolve from assistants to autonomous actors, technology leaders must rethink traditional governance models. The conversation can no longer focus solely on what AI can do. It must focus equally on what AI should be allowed to do.
This is where many organizations are beginning to converge around a common principle: innovation without governance creates risk.
Across the industry, leading technology organizations are emphasizing policy-driven AI, human oversight, explainability, and auditable decision-making as foundational requirements for enterprise AI. For example, Kyndryl’s Agentic AI Framework is built around the concept of securely orchestrating AI agents within defined governance boundaries rather than allowing unrestricted autonomy. The framework emphasizes policy enforcement, human-centric governance, auditable workflows, and enterprise-scale trust.
From my perspective as a CIO, this is exactly the right conversation for anyone considering agentic AI.
Healthcare organizations operate in one of the most highly regulated and mission-critical environments in any industry. Patient data, clinical decision support, financial transactions, and operational workflows require accountability at every step. As AI agents become capable of initiating actions, making recommendations, and orchestrating complex processes, governance can no longer be viewed as a compliance exercise. It becomes a prerequisite for adoption.
The organizations that realize the greatest value from agentic AI will likely be those that establish trust frameworks early, incorporating security, identity management, monitoring, and human accountability into the design process from the beginning.
In Part 2, I will cover what to put in place: least privilege, agent identities, and human oversight.
Let’s Connect
Talk with Healthcare IT Leaders about AI readiness for your health system. We can help you plan governance before agents go live.
Healthcare only works with Humans in the Loop.