Key Takeaways
- Healthcare was the second‑most targeted sector for ransomware in 2025, representing 22 % of disclosed attacks and affecting roughly 57 million individuals.
- AI‑related risks in healthcare often stem from uncontrolled use of generative AI (GenAI) with protected health information (PHI), clinical content, credentials, or internal workflows rather than from malicious model attacks.
- Organizations must shift from breach‑prevention to resilience: assume compromise, ensure rapid recovery, and maintain minimum viable operations during cyber events.
- GenAI amplifies classic attack vectors (phishing, social engineering, malware) and creates new risks such as prompt injection, data leakage, model manipulation, hallucinations, and insecure retrieval‑augmented generation pipelines.
- Effective AI governance requires a full inventory of AI systems, risk‑tiering by use case and data sensitivity, human‑in‑the‑loop controls for high‑risk applications, comprehensive logging, and rigorous vendor due‑diligence.
- Emerging regulatory updates—including the proposed HIPAA Security Rule, HHS guidance, HITRUST, PCI DSS, and FDA expectations for AI/ML‑enabled devices—are turning compliance into a resilience enabler that drives investment, testing, and repeatable processes.
- CISOs must translate AI risk into business terms (patient safety, downtime, trust, legal exposure, recoverability) and demonstrate to boards where AI is used, who owns it, how it is governed, and how it will operate during an incident.
The Evolving Threat Landscape in Healthcare
Healthcare remained one of the most frequently targeted industries for ransomware in 2025, accounting for 22 % of disclosed attacks in a widely cited analysis. Over the past year, 93 % of healthcare organizations reported at least one cyber incident, and by year‑end at least 642 large breaches had been disclosed, impacting nearly 57 million individuals. While these figures underline the volume of traditional cyber risk, the sector’s unique nature—where technology directly influences patient safety—means that any security failure can quickly become a care‑delivery failure, not merely a privacy breach.
How Generative AI Is Redefining Risk the rapid democratization of generative AI (GenAI) is reshaping both the threat landscape and the operating model in healthcare. Low‑cost or open‑source GenAI tools enable attackers to launch sophisticated campaigns faster and cheaper, using techniques such as “vibe coding” to craft convincing phishing lures, deep‑fake impersonations, and automated malware. At the same time, well‑meaning employees often unintentionally expose PHI, clinical data, credentials, or internal workflows by plugging sensitive information into unsanctioned AI applications, making uncontrolled AI use the first and most common AI problem in healthcare settings.
Expanded Attack Surface and New Vulnerability Classes GenAI introduces risk across multiple layers. Data leakage—especially PHI exposure—remains a top concern, while prompt injection and model manipulation can subvert AI outputs to malicious ends. Integration risks arise as AI copilots and agents connect to electronic health records (EHRs), collaboration platforms, knowledge bases, and ticketing systems, creating identity and authorization drift. The supply chain also widens: foundation‑model providers, AI plugins, model gateways, and embedded AI in SaaS services can introduce insecure retrieval pipelines, retrieval‑augmented generation (RAG) connectors, and unknown model or data provenance. AI‑generated hallucinations further threaten clinical decision‑making, amplifying patient‑safety concerns.
Principles for Safe GenAI Adoption To harness GenAI’s benefits while limiting exposure, organizations should undertake a disciplined approach: maintain a full inventory of all AI systems, models, agents, data sources, and connectors; risk‑tier each asset based on use case, data sensitivity, and level of autonomy; enforce human‑in‑the‑loop oversight for higher‑risk scenarios; log prompts, outputs, tool usage, data access, and administrative changes; and conduct thorough vendor due‑diligence for model providers and AI‑enabled software. These controls create a foundation for visibility, accountability, and rapid response when anomalies arise.
From Security to Operational Resilience The prevailing mindset for healthcare CISOs is shifting: it is no longer sufficient to ask “Are we using AI?” Instead, the focus must be on where AI is making decisions, summarizing information, triggering actions, or touching regulated data, and whether those workflows can be governed and recovered safely during a cyber event. Boards are now probing recovery speed and the ability to sustain minimum viable operations if core clinical, administrative, identity, or AI‑enabled workflows are degraded. This shift reframes cybersecurity as a patient‑safety imperative rather than a purely technical concern.
Patient‑Safety Impacts of Cyber Disruptions Research links ransomware and major cyber incidents to measurable harm in healthcare settings. One economics study found a 34–38 % increase in adverse events for patients already admitted when an attack begins. Other work shows a material rise in in‑hospital mortality during ransomware episodes at affected hospitals. These findings underscore that downtime, delayed care, or diverted resources translate directly into worsened clinical outcomes, reinforcing the need for resilience strategies that protect not only data but also the continuity of care.
Translating AI Risk into Business Language For CISOs, communicating AI risk effectively means framing it in terms that resonate with executive leadership: patient safety, operational downtime, loss of trust, legal and regulatory exposure, and the organization’s ability to recover. Demonstrating where AI is deployed, who owns each system, how it is controlled, and how it will continue to function—or be safely shut down—during an incident builds confidence among boards and helps prioritize investments that safeguard both information and care delivery.
Compliance as a Catalyst for Resilience Emerging regulations are turning compliance from a checklist into a resilience driver. The proposed HIPAA Security Rule (December 2024) would mandate written procedures to restore critical systems and data within 72 hours, while HHS guidance, HITRUST, PCI DSS for payment environments, and FDA expectations for AI/ML‑enabled medical devices collectively raise the bar for cybersecurity posture. Additionally, legislative efforts such as CIRCIA’s proposed 72‑hour incident‑reporting window and 24‑hour ransom‑payment disclosure create explicit timelines that force organizations to test and validate recovery capabilities.
Leveraging Compliance Frameworks By aligning with these standards, healthcare entities can justify resilience investments, drive systematic testing of backup and recovery processes, and build repeatable, auditable resilience workflows. Compliance alone does not secure GenAI, but it enforces discipline around AI inventory, risk classification, data handling, human oversight, logging, vendor review, and recovery testing—core components of a robust AI‑risk management program.
Winning with GenAI: Governance Over Speed The organizations that will thrive in the AI era are not those that rush to adopt GenAI with minimal controls, but those that establish clear governance, enforce strong boundaries, maintain continuous oversight, and prove their ability to recover when things go wrong. The CISO’s role is to ensure that AI adoption proceeds safely and responsibly, preserving care delivery even when a breach or AI‑related failure occurs. In doing so, they protect patients, sustain trust, and position their organizations to innovate without compromising safety.