‹ Browse all contests

2026 Ministry of the Interior and Safety AI Analysis Model Development Competition: Integrated Notice

Ministry of the Interior and Safety

Overview

A combined guide to an AI analysis model competition addressing multiple public-sector problems.

DeepVoice/BlackBox allow models, APIs and external data from lawful public resources, but final evaluation is offline.

NaraJangteo uses only the designated Gemma4 and bge-m3 models; fine-tuning and external data are prohibited.

External LLM APIs are allowed only for self-labeling the supplied materials, not in submitted inference code.

Notice dates

Notice start date
2026-08-18
Notice closing date
2026-09-29

Calendar dates from the saved notice. They do not establish a deadline time, time zone or current acceptance status.

Eligibility

Korean nationals, individually or in teams of at most 5.

Multiple accounts or participation in multiple teams are prohibited.

Minors need a participation-consent form including legal-representative consent.

Organizer/operator employees and relevant affiliate employees, among others, are excluded.

Minors need participation consent including legal-representative consent.

Organizer/operator and relevant affiliate employees, among others, are excluded.

Entry under a company name is unverified.

Deliverables

Submit task-specific AI models through DACON for 3 tasks: vehicle-motion analysis in dashcam traffic-accident footage, AI deepfake-voice detection, and legal-violation detection in tender notices/documents.

DeepVoice/BlackBox: submit offline-reproducible Python code and models in submit.zip.

Limits: inference 60 minutes; installation 10 minutes; ZIP 10 GB and extracted contents 32 GB.

Round 2 teams submit training code, two HWP reports on model development and training-data composition, and all training data.

No report length limit.

NaraJangteo: fixed LLM inference code and static RAG assets.

Self-labeling requires reproducibility materials including generation code, prompts, model version and generated labels

Competition No.

236749: Deep Voice Crime Response AI Detection Model Competition.

236753: an AI competition for an intelligent model to analyze intentional accidents in dashcam footage.

Sub-competition 236754: an AI competition monitoring legal violations in NaraJangteo self-issued bidding notices.

Cautions and verification notes

The organizer states that model submissions closed on 9/29, so the 10/16 results announcement is not used as the deadline for new submissions.

The consent terms for the 3 contests state that entrants retain rights to their submitted ideas. However, copyright and the right to create derivative works in winning outputs that receive a certificate or prize money belong exclusively to the organizer. Use or provision to third parties without the organizer's prior written consent is prohibited.

This review covers only some eligibility, purchasing and rights provisions in the consent terms for the 3 individual contests. It does not verify the full consent terms, a separate winner agreement or their legal effect.

Do not combine AI/API restrictions across competitions. Distinguish permission to use external APIs during self-label generation for KONEPS from their prohibition during submitted inference, and from the internet-isolated final evaluations of the other pair of competitions.

Details of the 3 individual competitions are linked within 1 consolidated announcement. The component competitions and their prizes are not added again to the new-announcement count or the combined aggregate.

Sources

Check the original notice for current rules. English includes editorial renderings of stored notices and organizer names, not verified official English names. Original records are preserved. Excerpts may be incomplete; consult the linked sources for full conditions.