Government-Led AI Model Flags Jeonse Fraud Risk Before Contracts
A government-backed AI model can spot high-risk jeonse fraud groups before contracts are signed. The National AI Strategy Committee said it completed policy research on an "AI Jeonse Fraud Early Detection Model" and confirmed it can detect about 60% of high-risk cases in advance, even with limited data.
The move comes as pressure builds to protect tenants. Just last month, victims' groups urged swift passage of the revised Special Act on Jeonse Fraud.
What's New
The model is the product of a research effort that began in October last year, led by Professor Lee Yong-jae at UNIST. The team worked with the Ministry of Land, Infrastructure and Transport, the Financial Services Commission, the Korea Real Estate Board, and Korea Credit Information Services to build and validate a pilot.
Roughly 3 million jeonse contract records were combined with landlord credit data to predict fraud risk at the pre-contract stage-where decisions actually get made.
Data Use and Privacy
Personally identifiable information was removed, and analysis ran in a closed environment. The approach shows it's possible to generate practical risk signals without exposing sensitive personal data.
What Signals Matter Most
Financial indicators were more predictive than property features. The strongest variables included the landlord's loan size, interest rate levels, recent delinquency history, and use of non-institutional financial services.
The team noted that performance should improve as the scope and quality of available data expand.
Why It Matters for Policy
"We attempted to develop a model that proactively detects future jeonse fraud risks by learning past data patterns," said Professor Lee, adding that the work could serve as foundational input for future prevention policy.
Choi Yoo-sam, head of Korea Credit Information Services, said the model is meaningful because it introduces criteria that can be applied during the contract stage-when tenants and agents need clear guidance.
Path to Public Service
Relevant ministries and public agencies, including the Ministry of Land, Infrastructure and Transport and the Financial Services Commission, will discuss how to expand this research into an actual public service. Negotiations will continue on data sharing and integration for key elements like delinquency and registration information.
The government also plans safeguards to prevent "social scoring" side effects-where AI outputs could lead to unfair stigmatization of specific landlords.
What Government Teams Can Do Now
- Plan pilot deployments with clear guardrails: define use cases limited to pre-contract checks, publish criteria, and set simple escalation paths for edge cases.
- Tighten data governance: standardize data quality checks, retention limits, and auditing across participating agencies; document model assumptions and known limits.
- Build human-in-the-loop review: require human confirmation for high-impact decisions, and give tenants and landlords transparent reasons and a way to contest outcomes.
- Coordinate outreach: brief local governments, realtor associations, and tenant groups so the risk signals are understood and used correctly.
Bottom Line
The pilot shows early promise: meaningful risk detection using financial signals, privacy protections in place, and a realistic path to service. With expanded datasets and strong guardrails, this can become a practical tool to cut fraud exposure before contracts are signed.
If your team is upskilling for AI oversight, risk evaluation, or model auditing in the public sector, explore curated options here: AI courses by job role.
Your membership also unlocks:
