AI news ·
Atlantic Health CIDO says AI governance and security are the most critical issues facing health systems today
AI governance is now the top priority for health systems as AI shifts from passive insights to active clinical decision-making. Without oversight, the technology introduces patient safety risks, bias, and expanded cybersecurity exposure.
Health Systems Must Build AI Governance Now or Face Clinical and Regulatory Risk
Governing and securing artificial intelligence has become the single most critical issue facing health systems, according to Sunil Dadlani, chief information and digital transformation officer at Atlantic Health in Morristown, New Jersey. AI is moving from passive insights to active execution within clinical and operational workflows - and health systems unprepared for that shift will struggle.
The opportunity is clear. AI can reduce documentation burden, close care gaps, accelerate prior authorization and improve patient access. But without rigorous governance and cybersecurity, the same technology introduces clinical risk, amplifies inequities and expands the attack surface for breaches.
Trust at Scale Is the Hardest Problem
Atlantic Health has seen pilots deliver wins - automation that improves pre-procedure readiness and reduces preventable cancellations, for example. But pilots also expose failure modes that don't appear in controlled settings: hallucinated content in summarization, drift in models trained on changing clinical language and vendor tools that cannot explain where their data comes from.
Every new AI integration increases identity, API and data exfiltration risk, especially when third parties handle protected health information. "Where this is heading is toward AI accountability frameworks becoming as standard as HIPAA compliance," Dadlani said. "Health systems that build that governance infrastructure now will lead. Those that don't will face consequences - regulatory, reputational and clinical."
Enterprise Operating Model, Not Scattered Pilots
Atlantic Health treats AI as an enterprise operating model rather than a collection of disconnected experiments. The organization built a unified governance structure that includes clinical leadership, compliance, privacy, security and operations. Every use case has an accountable clinical owner, defined success metrics and a monitoring plan before it touches production workflows.
On the technical side, Atlantic Health consolidated AI deployments behind a secure integration layer with least privilege access, strong identity controls and vendor due diligence. The health system requires data lineage, model limitations, auditability and clear practices for handling PHI.
A concrete example is the rollout of patient-facing automation for procedure preparation. It reduced avoidable cancellations and call volume, but Atlantic Health only scaled it after validating scripts with clinical teams, embedding escalation paths to humans and instrumenting it with quality checks and error reporting.
The next phase is continuous evaluation - measuring safety, equity and outcomes over time. AI performance is not a onetime certification; it is a living operational responsibility. Atlantic Health is piloting a real-time model monitoring dashboard that flags performance drift before it becomes a clinical risk.
Five Recommendations for Health IT Leaders
Stop treating AI as purely an IT problem. It is a clinical, ethical, financial and operational challenge simultaneously. Assign accountable leadership: every AI deployment needs an executive sponsor, a clinical owner and a risk owner across privacy, compliance and security. If accountability is diffused, safety and value will be inconsistent.
Get uncomfortable with the ethics conversation early. Questions around algorithmic bias, patient consent for AI-assisted care and liability when an AI recommendation contributes to an adverse outcome are no longer hypothetical.
Invest in data foundation before AI ambitions outpace infrastructure. Health systems often chase sophisticated AI use cases while sitting on fragmented, ungoverned data. You cannot build a reliable intelligence layer on an unreliable data layer.
Redesign workflows, not just technology. Many organizations underestimate change management. Train clinicians and staff on how to use AI safely, when to verify outputs and how to escalate to humans. Measure adoption and outcomes the same way you would for any clinical quality initiative.
Build external partnerships deliberately. Work with academic medical centers, technology leaders in AI, peer health systems and regulators. The challenges are too consequential and too complex for institutional isolation.
Learn more about AI for Healthcare and Generative AI and LLM technologies that health systems are deploying.