evidoria

← Back to browse

Good practice Imported

South Korea's National Election Commission — AI-Assisted Deepfake Detection for the 2024 General Election

South Korea · Gwacheon · See the South Korea profile

Evidence: Observational / pre–post Top 32% 67/100 · Ask Evidence Copilot about this practice

Ahead of its April 2024 general election, South Korea's NEC ran a task force under a new deepfake ban and detected 129 illegal AI-generated campaign posts in a 19-day window; police separately piloted an 80%-accuracy detection tool trained on 5.2 million data points.

129
Illegal AI-generated campaign posts detected (29 Jan-16 Feb 2024 (19-day window))
1,077 in 268 teams
Contest participants (deepfake-detection tool)
5.2 million data points from ~5,400 individuals
Training data points for the detection model
~80%
Detection tool accuracy
5-10 minutes
Processing time per video
7 years imprisonment or 10 million won fine (~US$7,500)
Maximum penalty for election deepfakes
South Korea's National Election Commission — AI-Assisted Deepfake Detection for the 2024 General Election

Details

Maturity
Pilot
Promoter
National Election Commission (NEC) of the Republic of Korea
Period
Jan–Apr 2024
Keywords
elections, public safety, content moderation, AI

Context

In December 2023, South Korea's National Assembly amended the Public Official Election Act to ban deepfakes for campaigning, effective 29 January 2024 ahead of the 10 April 2024 legislative election, with penalties of up to seven years in prison or a 10 million won fine.

Objectives

Detect and deter AI-manipulated campaign content across social media and other online platforms during the election period.

Activities

The National Election Commission formed a task force to screen platforms for AI-manipulated campaign content. Separately, the National Office of Investigation and National Forensic Service ran a public contest (1,077 participants, 268 teams) to build detection software, producing a model trained on 5.2 million data points from about 5,400 individuals.

Results

Between 29 January and 16 February 2024, the NEC task force identified 129 posts that violated the deepfake ban. The contest-built model produced a probability score in five to ten minutes per video with roughly 80% accuracy, designed to inform investigations rather than serve as standalone evidence.

Conclusions

Analysts noted the NEC's monitoring capacity is limited relative to online content volume, so the 129 confirmed cases likely understate true prevalence, and the 80%-accuracy tool is explicitly advisory rather than conclusive.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Short (< 1 year)
Staffing & skills
NEC task force, National Office of Investigation, National Forensic Service

Conditions for success

  • A specific legal ban with defined penalties in force before the election window
  • Cross-agency collaboration between the election commission, criminal investigators and forensic scientists
  • A public contest structure to source detection-model talent and training data

Common failure modes

  • Monitoring capacity is limited relative to the volume of online content, likely undercounting true deepfake prevalence
  • Detection tool accuracy (~80%) is advisory only, not admissible as standalone evidence

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful