For Cybersecurity Awareness Month, our sister publication, Healthcare Innovation (HCI), has published an article by David Raths featuring insights from Brian Anderson, M.D., president and CEO of the Coalition for Health AI (CHAI), on the growing cybersecurity risks associated with agentic AI in healthcare, including patient data protection, accountability, and trust.

Read an excerpt of the article below.

CHAI CEO: Health systems must address cybersecurity issues around agentic AI

With October being Cybersecurity Awareness Month, it is fitting that health system leaders are starting to focus on how AI introduces hazards as well as benefits to their data ecosystems. In August we reported that the Coalition for Health AI (CHAI) had launched a Health AI Cybersecurity Work Group. More recently, Brian Anderson, M.D., CHAI’s president and CEO, spoke in more detail about some of the issues the work group needs to address. 

Speaking during a Healthcare Summit hosted by cybersecurity company Rubrik, Anderson noted that with AI agents doing complex multi-step tasks involving patient data, the level of risk for health systems is suddenly getting more pronounced.

Founded in 2022, the nonprofit CHAI now brings together nearly 3,000 organizations across health systems, life sciences, health tech, academia, government, and patient advocacy. As we reported earlier, CHAI’s 100-member Health AI Cybersecurity Work Group will collaborate through bi-weekly work sessions to develop a range of outputs, including defensive and offensive playbooks, a frontier AI cyber risk assessment tool, and more. The leadership council will oversee these efforts and ensure the resources align with the threat levels hospitals and health systems are actively facing.

CHAI noted that the release of Anthropic’s Mythos-class frontier models, including the public Fable variant on June 9, 2026, introduced a fundamentally changed cybersecurity threat and defense landscape for healthcare. These Mythos-class capabilities compress attack timelines, industrialize the exfiltration of protected health information and ransomware, while offering defenders unprecedented capabilities for vulnerability management and incident response, the organization said.

Anderson said the rapid pace of AI adoption in health systems has raised risk management issues. Boards of hospitals and health systems are eager to invest and are asking what's the next ambient listening use case. But the cybersecurity infrastructure risk leaders are saying with agentic tools, there's a whole different set of issues to consider. “That is the main reason we wanted to launch this effort now, because seeing how quickly ambient listening was adopted and now that we're in this agentic space — we’re talking about clinical use cases, we're talking about PHI, we're talking about direct patient engagement with semi-autonomous agents doing complex multi-step tasks,” he explained. “The level of risk is getting more pronounced, and there's a greater seriousness and significance to it. I think it is well past time for us to have this effort. I'm excited that we finally have launched it.”

He noted that when they started CHAI, it was with a focus on clinical, administrative and back-end workflows. This concept of cybersecurity resiliency and putting in place the infrastructure needed to protect PHI was sort of an afterthought, Anderson said. “We launched this in 2022. Why didn't we start with a cybersecurity working group then? It's now 2026, and we're launching it. But I think, it is a growing issue now, and it is front and center at all these health systems. When we launched this cybersecurity work group, we talked to a few CISOs at major health institutions and they all said yes. Since we've announced the work group, we have over 300 organizations that are interested in participating. Not all of them are going to be able to participate, but it's a clear sign that this is front and center and really important to a lot of people now.”

Read the full article at Healthcare Innovation