Healthcare organizations are struggling to keep pace with the spread of AI tools, particularly unsanctioned agentic AI that employees deploy without formal IT approval, according to a new survey from Imprivata. The findings, published September 15, reveal that 72% of healthcare leaders admit some AI tools or agents are operating outside official oversight, creating security risks that traditional identity and access models were not designed to handle.
The report, "The Agentic AI Trust Gap: Why Healthcare Needs Identity-Led Governance," is based on responses from 250 U.S.-based healthcare leaders. It shows that agentic AI adoption is already well underway - 28% of respondents said they have agentic AI in production, and another 44% are piloting AI agents. Eighty-eight percent expect AI agents to operate with some degree of autonomy across clinical and operational workflows.
The shadow AI problem
Unsanctioned AI - often called shadow AI - poses a distinct threat because agentic tools may rely on public AI platforms. These tools can potentially share proprietary code and patient data beyond a health system's security perimeter. Even sanctioned agentic AI is moving faster than many organizations' ability to monitor agent activity, the researchers found.
While 86% of leaders said they are "fairly confident" they can control and govern AI agent actions today, the high rate of unauthorized deployment tells a different story. One senior manager from a 500-to-749-bed hospital system described a close call in the report: "Our hospital previously experienced an anomaly where AI autonomously exported patient information in batches. We were only able to quickly pinpoint the risk thanks to audit logs. Therefore, we only dare to expand the deployment of AI once our monitoring system is mature."
Identity and access risks
AI agents often interact with multiple patient care and research systems, retrieve sensitive data, and execute workflows through API activity. This range of access makes it harder to maintain visibility and accountability, according to Dr. Sean Kelly, Imprivata's chief medical and growth officer.
"An AI agent should be treated as a governable digital identity, with access appropriate to its role, clear limits on what it can do, and a record of its activity that can be monitored and audited," Kelly said. Security ranked as a top-three concern for 57% of respondents, who cited excessive or unnecessary access by agents interacting with clinical systems as a primary governance risk.
Governance models lag behind adoption
Health systems take varied approaches to agent identity, the report notes. As agentic AI becomes more autonomous, researchers said organizations must place equal emphasis on permissions and authorized activities. The survey aligns with previous findings that AI use is increasing at the point of care, even as governance frameworks remain nascent.
Lily Liu, digital health division director at Western Health, addressed this gap during an educational session at the HIMSS26 APAC conference in August. She said about 800 staff members at the Australian health system were using unapproved AI. Liu advised organizations to regularly update their AI governance policies and described a five-step assurance process for evaluating whether a proposed AI use case is appropriate. For professionals building expertise in this area, structured AI for Healthcare training can help teams understand both the clinical potential and the security implications of these tools.
Why this matters for healthcare professionals
Security teams and clinical leaders share responsibility for AI governance, and the survey makes clear that current oversight models are insufficient. Kelly summarized the core challenge: "As AI agents act on behalf of clinicians and staff, organizations need to understand what those systems can access, what they're authorized to do and how their activity can be monitored and reviewed." For cybersecurity teams in healthcare, the rise of agentic AI means identity and access management must evolve - and fast. The skills required to audit AI agent activity and provision appropriate access are increasingly overlapping with traditional security disciplines. Professionals who want to close that gap can benefit from dedicated AI for Cybersecurity Analysts learning paths that address monitoring, access control, and threat detection in AI-augmented environments.
Your membership also unlocks:
