Korean companies test sandboxed agent controls as LG opens K-EXAONE access

Hyundai Motor Group and Korean security company AIM Intelligence are exploring Nvidia’s OpenShell runtime for enterprise agents, while LG AI Research is opening API access to its K-EXAONE-2.5 model. The announcements combine a model-access update with early work on controlling what autonomous software can do.

OpenShell is a sandboxed runtime with policy controls for data, enterprise tools and agent actions. Hyundai is evaluating it alongside Nvidia’s NemoClaw blueprints. AIM Intelligence is integrating OpenShell with Starfort, its platform for detecting malicious instructions and data leakage. The companies are considering a preliminary proof of concept for the public energy sector, so this is not yet a reported production deployment.

LG AI Research’s K-EXAONE-2.5 was developed using Nvidia Nemotron datasets, including Nemotron-Personas-Korea. That dataset contains seven million fully synthetic Korean personas based on official statistics. Such records can help builders train or test locally relevant behavior without using real customer profiles, although synthetic data can still reproduce assumptions and gaps in its source distributions.

The releases address two different barriers to enterprise agents: access to regionally focused models and limits on tool execution. Whether the security combination works will depend on tests against prompt injection, unauthorized data access and policy bypasses under realistic workloads.