South Korea's Agency for Defense Development said on July 6 that Add+i, an in-house generative AI service, is now available to employees on its closed internal network, according to UPI and Korean press coverage. The system was built by the agency's Defense Artificial Intelligence Technology Research Institute using open AI models over about three months after a May 2025 task force decision. For practitioners, the important pattern is secure deployment architecture: document search, summarization, translation, and coding assistance can be useful inside defense R&D only when the model stack, retrieval data, update process, and logs stay within classified-network controls. The agency also says Add+i has already supported a geospatial information management system.
Add+i matters as an implementation pattern for sensitive AI adoption: it shows a defense lab choosing a closed-network generative system instead of sending research and administrative data to public services. The important question is not whether the features resemble commercial chat tools, but whether the deployment keeps data, retrieval, coding assistance, updates, and logs inside controls that match defense R&D risk.
What happened
UPI reported that South Korea's Agency for Defense Development launched Add+i, an in-house generative AI service for employees on the agency's closed internal network. Korean press coverage based on the agency announcement says Add+i was developed by the Defense Artificial Intelligence Technology Research Institute using open AI models after a May 2025 AI task force decision and about three months of development beginning in December.
Security context
The agency framed the service as a response to the defense sector's inability to use public generative AI tools because of military-secret leakage risk. Reported functions include internal regulation search, document summarization, translation, an AI development agent for closed-network work, and coding assistance for generating, testing, and debugging code. UPI also reported that Add+i has been used to build a geospatial information management system.
For practitioners
The transferable lesson is architecture discipline. Secure generative AI in classified or air-gapped settings depends on model provenance, retrieval boundaries, prompt and output logging, software-development controls, and offline update processes. Open models can reduce external dependency, but they do not remove the need for evaluation, access control, and auditability.
What to watch
The next useful signals are whether ADD discloses model-evaluation practices, guardrails for coding-agent output, data-retention rules, and how Add+i 2.0 expands tool access without weakening network isolation. Until those details are public, the safest assessment is a notable secure-AI deployment, not proof that the model stack is broadly reusable elsewhere.
Key Points #
- 1ADD launched Add+i on a closed internal network to give defense researchers generative AI tools without public-service data exposure.
- 2The system supports internal search, summarization, translation, coding assistance, and an AI development agent for closed-network workflows.
- 3Practitioners should focus on model provenance, retrieval boundaries, logging, and offline update controls before copying the pattern.
Scoring Rationale #
This is a notable secure-AI deployment because it shows a defense research agency putting generative AI and coding assistance inside a closed-network environment. The impact is limited by sparse public technical detail and a single-agency scope, so it stays below major industry-level impact.
Sources #
Public references used for this report. Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
[Active Search Campaigns by BudgetEasy](/problems/sql/active-search-campaigns-by-budget)
[High CPC Clicks & Poor Landing PagesMedium](/problems/sql/high-cpc-clicks-poor-landing-page)
[Campaign ROAS by Attribution ModelHard](/problems/sql/campaign-roas-by-attribution-model)
250 free problems · No credit card