• Healthcare 150
  • Posts
  • AI in Hospitals Meets the Agentic Economy: The Cybersecurity Reckoning

AI in Hospitals Meets the Agentic Economy: The Cybersecurity Reckoning

AI in Hospitals and the Cybersecurity Implications of the Agentic Economy

Author

Joaquin Astorga
October 19, 2026

Artificial intelligence is entering hospitals at the same time healthcare’s digital attack surface is becoming more complex, interconnected, and difficult to defend. That convergence will define the next phase of hospital technology investment.

The opportunity is substantial. The AI in healthcare market is projected to expand from approximately $37.0B in 2025 to $613.8B by 2034. That scale points to something larger than another software cycle. Healthcare is beginning to redesign parts of its operating architecture around systems capable of interpreting data, generating content, supporting decisions, and increasingly executing tasks.

Generative AI represents a smaller but rapidly expanding layer of that market. The segment is projected to grow from $2.6B in 2025 to $39.7B by 2034. The significance is not simply the growth rate. Generative interfaces are lowering the friction required for clinicians and administrators to interact with increasingly sophisticated AI systems.

The central question is therefore changing. Hospitals are moving beyond whether they should use AI and toward how deeply AI should be embedded into operations, and how those systems should be governed once they can act across clinical and administrative workflows.

Hospitals Are Moving From Experimentation to Infrastructure

AI adoption across healthcare remains uneven. MedTech currently leads at approximately 45%, followed by digital health, consumer health, and services at roughly 40%. Pharmaceutical products sit near 35%, while payers, pharmacies, and inpatient and outpatient care remain closer to 30%.

The differences are instructive. AI is moving fastest where workflows are already digital, data is relatively structured, and implementation can generate measurable operational value. Hospitals operate in a more difficult environment. Their technology estates combine legacy infrastructure, clinical systems, connected devices, administrative software, and third party applications accumulated over decades.

That complexity explains why hospital AI adoption can look simultaneously advanced and immature.

Only 18.7% of hospitals in the underlying functional data report AI adoption across at least one category. Workflow optimization reaches 12.9%, while routine task automation reaches 12.0%. Predicting patient demand and staffing requirements remain below 10%.

Hospitals are therefore not automating the enterprise in one move. They are selecting narrow workflows where AI can remove friction without immediately transferring broad operational authority to software.

Scale, however, changes the equation.

In 2024, AI adoption reached 96% among hospitals with more than 400 beds, compared with 80% for hospitals with 100 to 399 beds and 59% for facilities with fewer than 100 beds. Larger institutions have more resources to support implementation, governance, integration, procurement, and cybersecurity.

The resulting gap could become structurally important. Large health systems can increasingly test and govern AI internally, while smaller hospitals may rely more heavily on external platforms offering packaged capabilities.

That dependence creates two effects. First, turnkey platforms may accelerate access to sophisticated AI for institutions without large internal technology teams. Second, they increase reliance on third parties for critical workflows, concentrating operational and cybersecurity risk outside the hospital itself.

In the agentic economy, hospital size may therefore determine not only how quickly an institution adopts AI, but how much control it retains over the infrastructure supporting that AI.

Hospitals Are Buying Productivity Before Moonshots

The current priorities of hospital executives offer a clearer indication of where AI is headed.

Caregiver burden and satisfaction are identified as a top or second priority by 72% of respondents. Patient safety and quality follow at 58%, with workflow efficiency and productivity at 53%. Margin improvement ranks substantially lower at 12%, while patient and consumer experience reaches only 5%.

The hierarchy matters. Hospitals are not primarily implementing AI because they expect algorithms to produce immediate margin expansion. They are implementing it because healthcare has a capacity problem.

Clinicians remain overloaded by documentation, inbox management, scheduling, administrative tasks, patient communications, and increasingly complex information environments. AI can generate meaningful value by removing work around clinical care rather than replacing the clinical decision itself.

That is why practical applications are moving faster than broad autonomous systems.

Imaging and radiology lead deployment at 90%, followed by early detection of sepsis at 67%, ambient notes at 60%, clinical deterioration risk at 56%, unplanned admission risk at 52%, in basket automation at 51%, and medical coding at 45%.

These applications provide a roadmap for the next phase of hospital AI. The first generation largely analyzed data, generated documentation, prioritized information, or supported a specific decision. Agentic AI goes further.

A generative system can draft a discharge summary. An agent could potentially retrieve the relevant data, draft the summary, identify follow up requirements, schedule an appointment, notify the appropriate care team, prepare patient instructions, and update other systems involved in the workflow. That difference dramatically expands both the economic potential and the security consequences.

The Agentic Economy Expands the Attack Surface

Agentic AI changes the cybersecurity equation because useful agents require authority.

An AI system that only summarizes information needs access to data. An AI system that executes workflows may need permission to retrieve records, communicate with clinicians, schedule appointments, interact with billing systems, access pharmacy information, change administrative records, or trigger actions in other software.

The more useful the agent becomes, the more access it may require. Traditional hospital cybersecurity largely focuses on controlling employees, applications, devices, and external vendors. The agentic economy introduces another category of identity: autonomous software operating with delegated authority.

Hospitals will therefore need to define what each agent can see, what it can modify, which applications it can communicate with, which decisions require human approval, and how actions are logged for review. The risk is no longer simply that AI generates an incorrect answer. A poorly configured or compromised agent could potentially execute an incorrect action at machine speed. That makes identity, permissioning, and auditability strategic infrastructure.

AI Is Arriving Into an Existing Cybersecurity Crisis

Healthcare is not entering the agentic transition from a position of digital strength. The median number of individuals affected by a healthcare data breach increased from 2,030 in 2015 to 7,614 in 2022. It remained elevated at 7,362 in 2023 and 6,146 in 2024, before falling to 4,011 in 2025 and 3,651 in 2026 in the supplied data.

The recent decline is encouraging, but it does not eliminate the structural problem. Healthcare organizations hold unusually sensitive information, operate complex technology environments, and cannot tolerate downtime as easily as most commercial enterprises.

The largest breaches demonstrate another vulnerability that becomes more important as AI ecosystems expand: concentration risk.

The Change Healthcare incident affected approximately 192.7M individuals. Anthem’s 2015 breach affected approximately 78.8M. Multiple other incidents affected more than 10M individuals.

The pattern matters because hospitals will rarely develop every component of their AI infrastructure internally. Deployment will depend on EHR vendors, cloud infrastructure, AI model providers, cybersecurity platforms, medical device manufacturers, workflow software, and specialized data companies.

Third party risk therefore becomes part of AI risk.

An integrated platform creates efficiency precisely because it connects previously separate systems. But deeper integration also means a successful compromise can potentially propagate across a broader set of workflows.

Ransomware Makes AI Resilience a Clinical Issue

Ransomware illustrates why hospital cybersecurity is fundamentally different from cybersecurity in most industries.

Approximately 77% of healthcare organizations were targeted by ransomware in the prior year according to the supplied data, while 67% were hit. Healthcare accounted for approximately 17% of ransomware attacks, the highest share attributed to any industry in the chart.

Among affected organizations, 53% paid ransom. The average ransom demand reached $7M, while the highest recorded demand reached $100M. Average recovery cost stood at approximately $2.57M

For a conventional enterprise, ransomware can stop business operations. For a hospital, it can interfere with care delivery.

Scheduling can become unavailable. Imaging systems can be disrupted. Records may become inaccessible. Administrative functions can stop. Clinical teams may be pushed onto manual processes at precisely the moment when speed and information matter most.

Greater AI dependence adds another dimension to resilience planning.

Hospitals must not only ask whether an AI system performs effectively under normal conditions. They must ask what happens when the AI system, its model provider, its data connection, or the underlying workflow becomes unavailable.

Agentic healthcare therefore requires fallback pathways. Critical processes should degrade safely rather than fail completely.

Medical Devices Expose the Hardest Part of the Problem

The collision between AI adoption and medical device security may be the most difficult challenge hospitals face.

Imaging and radiology are currently among the most mature hospital AI use cases. At the same time, medical device infrastructure remains heavily exposed.

According to the supplied data, 99% of hospitals operate devices containing known exploited vulnerabilities. 93% have known exploited vulnerabilities combined with insecure internet connections for connected medical devices. 89% have devices containing vulnerabilities linked to ransomware.

Most importantly, 85% of organizations manage imaging systems containing these vulnerabilities.

This creates an uncomfortable contrast.

Some of the hospital environments moving fastest toward AI are also operating on top of some of healthcare’s most difficult legacy infrastructure. Connected devices can remain in service for many years. Operating systems may become unsupported. Credentials may be weak or difficult to change. Patching can require coordination with manufacturers and clinical teams because downtime itself carries operational consequences.

AI modernization therefore does not automatically modernize the technology beneath it.

Connecting an intelligent layer to vulnerable infrastructure can create significant productivity gains. It can also increase the consequence of compromise.

Cybersecurity Becomes Part of the AI Investment Thesis

For hospital executives, the next phase of AI requires a different procurement framework.

Model performance will matter, but it will not be sufficient. Hospitals will increasingly need to evaluate permission controls, identity management, auditability, data governance, network segmentation, monitoring, incident response capabilities, third party dependencies, and the ability to limit autonomous actions.

The safest agent will not necessarily be the least capable agent. It will be the system whose capabilities can be clearly bounded.

For investors, cybersecurity therefore moves directly into AI underwriting.

A hospital AI company may appear attractive because its automation produces measurable labor savings, improves clinical capacity, or embeds itself deeply into workflows. But deeper integration also increases the importance of the vendor’s security architecture.

This creates a meaningful distinction between the first generative AI cycle and the emerging agentic one.

Generative AI companies largely competed on model capability, user experience, integration, and accuracy. Agentic healthcare companies will increasingly compete on controlled autonomy.

An AI platform that can act across an EHR, scheduling system, revenue cycle platform, communications layer, and clinical workflow potentially becomes extraordinarily valuable infrastructure. It may also become extraordinarily difficult to replace.

That combination can create strong retention and attractive economics. But it raises the cost of failure.

Hospitals therefore do not need the most autonomous AI available. They need the most autonomous AI they can safely govern.

The same applies to investors. The strongest opportunity may not be the company promising to remove humans from every workflow. It may be the company capable of automating enough of the workflow to produce measurable value while maintaining clear boundaries around what software is allowed to do.

Healthcare is moving from software that helps people make decisions toward software that increasingly participates in those decisions and executes portions of the surrounding workflow.

The productivity opportunity is significant. So is the operational exposure.

For hospital executives, clinicians, technology vendors, and investors, the strategic conclusion is becoming difficult to avoid.

AI is becoming hospital infrastructure. Cybersecurity must become infrastructure at the same speed.

Sources & References