Korean Startup Siseon AI to Build Coding AI for Air Force's Offline Network

South Korean startup Siseon AI has signed a contract with the Republic of Korea Air Force to deploy an artificial intelligence platform that will write and check code inside a defense network cut off from the internet. It is one of the first cases in which generative coding AI has been deployed to a system running in full network isolation rather than in the cloud.

What the deal covers

The contract covers building a platform that combines two functions: a coding assistant and an agent that automatically checks the generated code for security vulnerabilities. The whole system will operate under the government program "Support for Rapid Commercialization of AI Products in the Defense Sector," overseen by IITP, an agency under South Korea's Ministry of Science and ICT. The end user is the Republic of Korea Air Force.

Siseon AI is not running the project alone. The team includes Sejong University's Industry-Academic Cooperation Foundation, responsible for research and technology verification, and CodeMind, a company that supplies automated source-code vulnerability detection technology.

Why an offline network

Developing military software requires boosting both productivity and security at the same time, without letting sensitive data or source code leave the organization. Cloud-based generative AI models can speed up developers' work, but their deployment is limited in environments that require network separation and strict control over information leaks.

That's why the key task becomes AI running on-premise, on servers physically cut off from external networks, combined with technology that automatically identifies vulnerabilities in generated code. IntraGenX is designed to address exactly this problem: it operates in a closed environment and uses a lightweight 30-billion-parameter language model to generate and evaluate code.

How IntraGenX works

The platform is built on a structure called "Loop Harness," which iteratively evaluates and refines the model's output. On top of that sits a knowledge-graph-based RAG engine that analyzes relationships between source code, database schemas, military development guidelines, and unstructured documents. The system tracks calls, references, and dependencies between code and data, presenting the reasoning behind its output, not just the code itself.

This contract matters because it lets us verify IntraGenX's security and technological capabilities within a defense network, and it secures a key market reference for entering this space - Nam Woon-sung, CEO of Siseon AI

Ambitions of a Korean Palantir

After verifying the technology within the Air Force network, the company plans to assess its potential use in the Army, the Navy, agencies directly under the military, and the Ministry of National Defense itself. Siseon AI also talks about ambitions to build a strategy comparable to America's Palantir, the company known for data analysis work for the Pentagon, the CIA, and the FBI, by structuring the data and operational knowledge of high-security organizations into ontologies that AI can operate on.

The company says that after defense, it wants to move into the financial sector, large corporations, and defense manufacturers, all of which require strict security regulations and network separation. It's a structurally similar group of clients, since similar restrictions on moving data outside controlled environments apply across all of them.

What it means for the market

For Polish companies and institutions grappling with how to deploy AI coding assistants without risking sensitive data leaking to cloud model providers, this case points to one possible direction: on-premise models combined with automated code security checks, instead of giving up on AI altogether. The question is especially relevant for regulated sectors where network separation is a legal requirement, not just good practice.

The contract's value, 2.45 billion won, is a relatively small pilot by the scale of defense budgets, but for a startup it serves as a reference that could open the door to much larger orders in the military, financial sector, and defense industry. For now the project covers only the Air Force, and its further expansion depends on how the technology performs in a real production environment.