AI’s defensive power in real time
Greenhalgh also spotlighted how AI can help defenders gain the upper hand. In healthcare, he said, it’s proving most effective for real-time anomaly detection and predictive threat intelligence.

“Healthcare systems generate vast amounts of data—network traffic logs, access patterns, device communication,” he said. “AI, particularly machine learning models, can continuously analyze this massive data stream to identify deviations from normal behavior, even subtle ones that human analysts might miss.” That capability translates into tangible benefits: AI can detect suspicious login behavior, irregular data movement, or unusual device communication — red flags that suggest a possible attack and demand immediate attention.
“Furthermore, by learning from past attack patterns, AI can predict potential vulnerabilities and proactively recommend security enhancements,” Greenhalgh said.

Edge AI: Opportunity — and complexity
To date, the U.S. FDA has greenlit more than 1,000 AI-enabled medical devices, with the vast majority in radiology. That number is only expected to climb, Greenhalgh told AAMI attendees. “In the next two to three years, HTM and IT leaders need to be keenly aware of the accelerating trend of AI at the edge within medical devices and connected systems.”

Edge AI — AI that runs directly on medical devices rather than in the cloud — is reshaping healthcare, from wearables to imaging systems and beyond. The benefits, Greenhalgh said, are clear: faster responses and stronger privacy protections. But those advantages bring added complexities.

For starters, Greenhalgh asked, “How do we securely update AI models on thousands of distributed devices?” Then there’s the question of how to ensure the integrity and provenance of edge-generated data. “And critically, how do we manage the complex interoperability requirements between these intelligent edge devices and the central healthcare IT infrastructure, especially concerning data flow and security protocols?”

The responsibility, he added, increasingly falls on HTM professionals. These teams already manage device maintenance. But with a surge of FDA-cleared AI-enabled devices, are they now on the hook for AI, too?

Greenhalgh also cautioned against the assumption that AI systems are inherently secure. AI isn’t magic; it’s code, and like any code, it can be exploited. Bias, poisoning, adversarial inputs — all of these are real risks, he told AAMI attendees.

HTM Homepage