Responding to “shadow AI” with data access controls and sandbox isolation
Step-by-step process for advancing the generative AI PoC at Golden Planet and Eunpyeong St. Mary’s Hospital. Source: Golden Planet
Nurses, pharmacists, and administrative staff at The Catholic University of Korea Eunpyeong St. Mary’s Hospital (Director Bae Si-hyun) directly planned and developed AI agents for work tasks. Cloud company Golden Planet (CEO Kim Dong-sung) announced on the 3rd that it had released the results of a four-month generative AI proof-of-concept (PoC) project conducted with Eunpyeong St. Mary’s Hospital.
The two parties held the “Eunpyeong AX Summit 2026” at the hospital’s main auditorium on 29 September and presented the project outcomes. Based on Google Cloud’s enterprise AI platform “Gemini Enterprise,” frontline staff identified areas for improvement and directly created tools to address them. Golden Planet was responsible for training and technical support.
Training was conducted in stages. A basic no-code course was first offered so that faculty and staff with no development experience could participate, followed by hands-on practice in agent development. When the implementation group “AI Beta Crew,” formed through an internal idea contest at the hospital, derived project tasks, Golden Planet engineers provided one-on-one advice on prompt design and development direction. Among these, some agents with high expected usage were selected for PoC to verify their applicability to actual work.
Security issues were addressed in parallel. To block so-called “shadow AI,” in which hospital information is entered into unauthorized external AI services, an architecture based on “Harness,” which controls data access permissions, was applied. According to Google Cloud, Gemini Enterprise allows administrators to manage user permissions and activity history, and does not use input data for AI model training.
The operating environment was also segmented. A sandbox was created where anyone can experiment, and a separate certified area was designated where only agents that have passed security and personal data verification are officially used.
Based on this case, Golden Planet plans to expand its AX business in the medical and public sectors, bundling infrastructure deployment, AI training, security design, and PoC development.
CEO Kim Dong-sung said, “This is a case that demonstrates how hospital frontline staff can directly identify on-site issues and improve them using AI technology,” adding, “We will support medical institutions and companies, where data security and governance are of critical importance, in safely adopting AI.”
ⓒ dongA.com. All rights reserved. Reproduction, redistribution, or use for AI training prohibited.
Popular News