Healthcare AI is entering a new era of risk, and leaders are now confronting a challenge they did not anticipate when the technology first arrived: how to secure it. October marks Cybersecurity Awareness Month, and the sense of urgency has sharpened considerably. The Coalition for Health AI (CHAI), a nonprofit launched in 2022, brings together nearly 3,000 members from health systems, life-science firms, and government agencies. To address the emerging threats posed by agentic AI—tools that can act autonomously and perform multi-step tasks involving patient data—CHAI has assembled a 100-member cybersecurity work group. The group’s charter emphasizes protecting protected health information while enabling innovation across the sector.
The shift began with the release of Anthropic’s Mythos-class frontier models, including the public Fable variant in June 2026. These systems did more than accelerate AI capabilities; they altered the cybersecurity environment. Attack timelines collapsed from days or weeks to milliseconds or seconds, turning traditional defenses into liabilities. Exfiltration of protected health information (PHI) and ransomware attacks became industrialized, while defenders gained new tools for vulnerability management and incident response. CHAI noted that the rollout, which debuted the public Fable variant on June 9, 2026, fundamentally reshaped both the threat vectors and defensive postures for healthcare organizations.
Brian Anderson, M.D., CHAI’s president and CEO, framed the issue during a Healthcare Summit hosted by Rubrik. He explained that the problem is not only the speed of attacks but also the autonomy of AI agents operating inside clinical and administrative workflows. Hospitals and health systems are eager to adopt these tools, yet the risks now sit front and center on executive agendas. Boards are asking about “ambient listening” use cases, but the deeper concern involves agentic systems performing complex tasks such as direct patient engagement, PHI handling, and semi-autonomous decision-making.
The work group’s first challenge is straightforward: unintended data leaks. Health systems want assurances that PHI remains secure, yet frontier AI developers are reluctant to accept full accountability. Contracts stall over liability frameworks, and many health systems remain hesitant. Anderson noted, “Many of the health systems are shy right now and are stuck in the contractual place.
They want to have an enterprise relationship with these big AI companies, but they can’t come to terms on an agreeable framework for accountability and liability in an agentic workflow.” The 100-member group meets bi-weekly to produce concrete deliverables such as defensive and offensive playbooks, a frontier-AI cyber risk assessment tool, and additional resources designed to keep pace with the rapidly evolving threat environment.
Anderson acknowledged that cybersecurity was an afterthought when CHAI launched in 2022. The original focus centered on clinical and administrative workflows, with PHI protection treated as a secondary concern. By 2026, the situation had reversed, and security considerations now dominate strategic discussions. The work group’s rapid interest, over 300 organizations expressed participation, confirms the shift. Major health-system CISOs voiced the need for robust safeguards during early conversations, and the response since launch has been overwhelming.
Leave a Reply