As healthcare organizations move from AI experimentation to enterprise deployment, leaders are grappling with how to balance innovation with governance, safety and real operational impact.
At Atlanta-based Emory Healthcare, that responsibility sits with Nabile M. Safdar, MD, chief AI officer for Emory Healthcare and Emory University. A radiologist by training, Dr. Safdar oversees AI governance, portfolio management and performance evaluation across clinical, operational and patient-facing use cases.
“Radiology is one of the most digital specialties and one of the areas where early adoption of AI has been really prominent, to the point where many people have speculated that we could go away as a profession,” he said during an interview with the “Beker’s Healthcare Podcast.” “Well, I was interested in that, out of self preservation, and I thought I’d learn more.”
That curiosity evolved into hands-on work in AI research, implementation, deployment and governance, ultimately leading to his current role. Today, his team helps design governance for the system, manage the portfolio of AI efforts and evaluate performance.
AI needle-movers
Like many health systems, Emory has focused first on AI applications with clear regulatory pathways and measurable returns. Dr. Safdar said several familiar use cases are already delivering value.
“We’ve had a lot of success with ambient AI,” he said. “We’ve been using FDA-cleared AI in areas like cardiology and radiology and, hopefully soon, pathology as we do our digital pathology transition.”
Beyond clinician-facing tools, Emory is seeing early traction with patient outreach and remote monitoring through agentic AI. According to Dr. Safdar, these tools are beginning to close gaps in adherence and follow-up care.
“We see early success with agentic chatbots who are reaching out to patients to help remind them to make sure they fill their prescriptions and get their medications after discharge,” he said.
Those same agents are also supporting home monitoring workflows, with escalation built in when patients fall outside safe parameters. A strong example is blood pressure monitors; patients at risk can take regular blood pressure readings to ensure they’re in a safe range and if not, the technology will quickly escalate concerns to a human who can help them navigate the next steps.
Internally, demand for generative AI tools has grown rapidly as clinicians and administrators compare workplace technology to what they use in daily life.
“So many people, physicians, nurses, technologists, administrators, managers in healthcare are using something like ChatGPT, Claude or Gemini at home in their everyday lives,” Dr. Safdar said. “They want to have that same efficiency in their work as well.”
In response, Emory has deployed secure, HIPAA-safe environments that allow staff to use frontier models in appropriate ways, while broader opportunities in revenue cycle, marketing and access remain in earlier stages.
Avoiding governance extremes
As AI adoption accelerates, Dr. Safdar cautioned leaders against swinging too far in either direction on governance. In his view, both a lack of oversight and overly centralized control can derail progress, which becomes a bottleneck.
“Governance can be so restrictive that there is only a single committee which becomes a choke point in the system for the deployment of novel AI solutions,” he said. “Any large system needs to be more nimble than that.”
Emory’s approach pushes decision-making closer to the work. Functional areas and clinical specialties are encouraged to develop their own AI governance structures, with escalation triggers for higher-risk use cases. Those triggers include patient-facing autonomy, agentic behavior, conflicts of interest or unauthorized AI use. When triggers are present, decisions are elevated for additional oversight.
“They’re the experts that know what they need,” Dr. Safdar said. “We should trust them to make appropriate decisions, especially if we provide them with the triggers that would indicate a need for a higher level of scrutiny or review.”
Navigating a shifting regulatory landscape
Layered on top of internal governance is an increasingly complex regulatory environment. Keeping pace with AI policy requires coordination across multiple agencies and levels of government.
“You have to know the players,” he said, citing Congress, state legislatures, CMS, HHS, ONC and the Office for Civil Rights. “It can be a full time job to keep an eye on the regulations affecting digital and AI in a health care environment.”
Emory’s AI team works closely with regulatory affairs while tracking state-level activity, including hundreds of proposed or active AI bills nationwide. Dr. Safdar pointed to California’s SB 53 as one example that could shape how foundation model developers disclose safety practices and how health systems use those tools downstream.
“We’re waiting to see what kind of impact that’s going to have on us in healthcare and to those that use their products as derivatives through APIs, or through other means, and what it means when you’re trying to fine tune a model for healthcare,” he said.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.