Official AI Self-Regulation System for INS

AI That Judges
Editorial Norms

Norm + AI. An AI-based full-coverage review system that automates editorial self-review for online newspapers.

50+
Media Monitored
4.3s
Per-Article Analysis
$0
Extra Cost for Pledged Members
14
Review Categories
0
Articles Collected
0
Reviews Completed
23
Registered Media
129
INS Pledged Members

WHY NORMIA

Why AI-Powered Self-Regulation Is Needed

129 INS Pledged Members and Growing

129 media outlets have signed the INS self-regulation pledge. Managing this growing base with traditional methods is no longer feasible.

Tens of Thousands of Articles Daily

Tens of thousands of articles are published every day. Reviewing all of them with human reviewers alone is physically impossible.

Reader Trust & Quality Control Are Essential

Improving the credibility of online news requires systematic quality control. Self-regulation is the cornerstone of earning reader trust.

CURRENT LIMITATIONS

The Limits of Current Self-Regulation

01

Blind Spots of Random Sampling

Only a tiny fraction of all articles is sampled and reviewed. The vast majority of violating articles remain in the review blind spot.

02

Limits of Manual Review

Review committee members read and judge articles by hand. Daily throughput is limited, and fatigue-driven judgment errors are inevitable.

03

Lack of Consistency

Review criteria vary by individual. Different reviewers can reach different conclusions on the same article.

SOLUTION

How NORMIA Transforms the Review Process

BEFORE

Manual Review

  • Random Sampling — Only 1–2% of all articles reviewed
  • Manual Review — Minutes to tens of minutes per article
  • Limited Categories — Only major items are checked
  • Reactive Response — Action taken only after issues arise
  • Inconsistent Standards — Criteria differ between reviewers

AFTER — NORMIA

AI-Powered Full-Coverage Review

  • Full Coverage — Every published article is analyzed
  • 4.3-Second Analysis — AI processes each article in 4.3 seconds
  • 14 Categories — Comprehensive review criteria applied
  • Real-Time Detection — Analyzed immediately upon RSS collection
  • Consistent Standards — Judged by the same AI model throughout

PIPELINE

4-Stage Review Pipeline

Fully automated from RSS collection to dashboard reporting.

1

RSS Collection

Real-time monitoring of
RSS feeds from 50+ media

2

Tier-1 Filter

High-speed primary screening
via keywords and patterns

3

Tier-2 LLM

In-depth violation analysis
by Large Language Model

4

Dashboard

Violation stats & details for
committee member review

Tier-1

High-Speed Filtering

Regex and keyword matching quickly classify clear violation candidates. Processing time: under 0.1 seconds per article. Cuts unnecessary LLM calls and reduces cost.

Tier-2

In-Depth LLM Analysis

A Large Language Model reads the full article and judges violations across 14 categories, with contextual understanding and nuance detection.

Dashboard

Reviewer's Toolkit

Violation detection results are shown on a real-time dashboard. Committee members review the AI analysis and make the final judgment.

ADVANTAGES

NORMIA's Core Advantages

100%

Full Coverage Review

Every published article is analyzed, not just a sample. The review blind spot is eliminated at its root.

4.3s

Per-Article Analysis Time

AI analyzes a single article across 14 categories in an average of 4.3 seconds.

14

Review Categories

From Misinformation (A) to International Reporting Ethics (N) — 14 categories aligned with INS guidelines.

82%

Efficiency Gains

AI handles primary screening, freeing committee members to focus on cases that truly require human judgment.

$0

No Additional Cost

Provided at no extra cost to pledged members thanks to in-house infrastructure and an optimized LLM pipeline.

HITL

Human-in-the-Loop

AI is a screening tool. The final judgment is always made by a human review committee member.

14 CATEGORIES

Review Categories A — N

14 categories based on INS self-regulation guidelines.

A

Misinformation & Manipulation

Spreading inaccurate information

B

Balance

One-sided or biased reporting

C

Headline Appropriateness

Clickbait or exaggerated headlines

D

Source Attribution

Missing citation of sources

E

Sensationalism

Provocative language or imagery

F

Human Rights Violations

Privacy breaches or defamation

G

Discrimination & Hate

Group-based bias or hate speech

H

Crime Reporting

Glamorization or risk of copycat

I

Disaster Reporting

Disaster-reporting ethics violations

J

Children & Youth

Minor-protection violations

K

Conflict of Interest

Undisclosed ads or sponsorships

L

Copyright

Unauthorized reproduction

M

Reporting Standards

Violation of basic reporting principles

N

International Reporting Ethics

Violation of international reporting norms

Violation Detection — Last 7 Days (based on the live dashboard)

1,687
M Ad-Content Blending
117
D Headline Rules
101
A Accuracy
21
Other (N, L, B, H, E)

DASHBOARD

Real-Time Monitoring Dashboard

An intuitive dashboard for review committee members. See the full violation landscape at a glance and drill into any case instantly.

INS Self-Regulation Monitoring Live
Ollama Connected 23 RSS Feeds Active
Articles Collected
25,691
Reviews Completed
8,414
Violations Detected (7d)
1,926
Compliance Rate
95.0%
Violation Distribution by Category (7 days)
M Ad-Content Blending
1,687
D Headline Rules
117
A Accuracy
101
N AI Content
8
Other (L, B, H, E)
13
Real-time violation detection status and statistics
Violation distribution visualized across 14 categories
Original article and AI analysis rationale view
Severity (High / Medium) filtering for violations
Daily / weekly / monthly trend analysis
Committee member review and comment logging

PRINCIPLES

NORMIA's Three Principles

AI as a Screening Tool

Its role is to surface potentially violating articles to committee members. It does not issue automatic sanctions or final verdicts.

Humans Make the Final Call

All final violation decisions rest with the review committee. AI analysis is only a reference, never the verdict.

Context Over Keywords

LLM-based analysis understands the full context of an article rather than relying on simple keyword matching, minimizing false positives.

FAQ

Frequently Asked Questions

How accurate is NORMIA's review?
NORMIA achieves over 90% accuracy in major categories, and continues to improve through ongoing model training and feedback. Since final judgments are always made by review committee members, false positives never translate into real harm.
How long does it take to analyze a single article?
The Tier-1 high-speed filter takes under 0.1 seconds; including Tier-2 in-depth LLM analysis, the average end-to-end time is 4.3 seconds.
How does a pledged media outlet sign up?
After signing the self-regulation pledge through INS and registering an RSS feed, monitoring begins automatically. System inquiries: normia@ivan.ai.kr
Are actions taken automatically when a violation is detected?
No. NORMIA follows the HITL (Human-in-the-Loop) principle. The AI only surfaces potentially violating articles to the dashboard — it never issues automatic sanctions. Final judgment is always made by review committee members.
Are there any additional costs for pledged members?
No. Thanks to in-house infrastructure and an optimized LLM pipeline, the service is provided to pledged members at no additional cost.
Is API integration available?
Yes. We provide a REST API. Detailed capabilities can be reviewed on the dashboard, violation list, and technical documentation pages.

Experience NORMIA

Check the live analysis status on the dashboard.