Korean firms use NVIDIA OpenShell to make AI agents safe
OpenShell limits what an AI agent can touch, and AIM Intelligence joins it with its own detection of malicious instructions.
NVIDIA says that Hyundai Motor Group is exploring its NemoClaw blueprints and the OpenShell runtime. AIM Intelligence, a Korean AI security company, joins its Starfort agent control platform with OpenShell. LG AI Research opens API access for K-EXAONE-2.5. The goal of AIM Intelligence is to join the access controls of OpenShell with its detection of malicious instructions and data leaks. The source gives no test results, dates or prices.
What OpenShell does
OpenShell is a safe runtime. It operates agents in a sandbox (a closed space) and uses policy based controls. These keep sensitive data safe, control access to company tools and agent tasks.
Who does what
- Hyundai explores the tools for employee and company workflows.
- AIM Intelligence joins runtime access controls with detection of data leaks.
- LG AI Research made its model with NVIDIA Nemotron datasets.
What is not known
AIM Intelligence only thinks about a first proof of concept for the public energy sector. No firm gives results or a start date.
Insight: NVIDIA makes a full stack for agents
Earlier, Trend AI put Nemotron in customer data centers. NVIDIA added Confidential Bare Metal for model makers and made Run:ai open source. At this time, OpenShell and Nemotron data are in Korean plans. This reads as NVIDIA making a stack from hardware to agent safety. Each partner adds a part and depends on its layers. Tell vendors what lock in a safety tool brings.
Sources
Posted
