A Light-Touch System Turns 60
On June 30, 2026, the Korea Media Rating Board (KMRB) marked its 60th anniversary with a forum in Busan titled "Beyond Regulation, Towards Industry." Chair Kim Byung-jae used the occasion to make a claim regulators rarely make about their own history: that the board's biggest achievement wasn't a rule it enforced, but a rule it gave up. "The greatest innovation in the Board's 60-year history was the introduction of OTT self-rating classification," Kim said, adding that AI-based classification will be the next major shift (Sports Khan).
That self-rating system, codified in Article 50-4 of the Promotion of Movies and Video Products Act, lets designated OTT platforms — Netflix, Wavve, TVING, Coupang Play, and Disney+ among them — classify their own content against KMRB standards rather than submitting every title for pre-clearance (Korea Law Information Center). Operators report their ratings to the board, which retains ex-post authority to override a classification it disagrees with. It is, in regulatory terms, a bet that industry can be trusted to self-police if the state keeps a credible backstop.
The Volume Problem Is Real
The case for tightening that arrangement is not frivolous. At the forum, U.S. content-analysis firm Spherex demonstrated AI that scans video, audio, subtitles, and sound effects to flag context a human reviewer might miss across roughly 200 regions' worth of cultural norms — CEO Teresa Phillips noted that a model trained mainly on Western content can misread something as culturally specific as a Korean bathhouse scene (Herald Corp). Panelists at the forum argued that content volume is now outpacing what a fixed roster of human reviewers can credibly assess, and that stretched review capacity is exactly the condition under which a self-rating regime could quietly drift toward laxity — a platform grading its own homework has an incentive to rate leniently, not conservatively.
The data gives that worry some traction. Since the self-rating system took effect in June 2023, the share of OTT content rated 19-and-over fell from 20.6% to 14.1%, while all-audiences content rose from 21.7% to 40.8% (ETNews). A skeptic could read that as ratings loosening precisely as oversight capacity is under strain — a fair argument for firmer intervention: mandatory algorithmic audits, faster override powers, or stiffer penalties for mis-rating.
Why AI-NI, Not AI-Only, Is the Right Design
But the same ETNews data cuts the other way on the actual question KMRB is answering. Direct reclassifications — the board stepping in to overrule a platform's self-rating — totaled six cases (2.9%) in 2023 and just two (1.1%) in the first three quarters of 2024, with operators self-correcting an additional 43 times (23.8%) without formal enforcement. A system correcting itself in under 2% of cases is not a system in crisis; it's a system whose marginal problem is throughput, not accuracy. That distinction is why KMRB's answer — an "AI-Natural Intelligence" model in which AI analyzes content first and a human expert makes the final call, rather than an AI making the call outright — is the proportionate response. It scales the board's ability to keep pace with volume without swapping a working, human-adjudicated system for an algorithmic one whose cross-cultural blind spots (the bathhouse example again) are exactly what a below-2% error rate would start to erode.
Soft-Law Incentives Are Doing More Work Than Mandates
KMRB's other lever this year has been reputational, not punitive. On March 11, 2026, the board named TVING and Disney+ its "Outstanding OTT Self-Rating Operators" for 2025, evaluated against eight other platforms. TVING — the first domestic operator to win the designation — was cited for a five-step internal review process and a granular 7+ age tier; Disney+, honored for a second consecutive year, was recognized for consumer surveys, a college-student monitoring group, and proactive content warnings (Sports Khan). Kim Byung-jae framed the award as recognizing operators moving "beyond mere legal compliance." A public dataset of self-rating decisions, published on Korea's open-data portal and updated as classifications come in, adds a further layer of ambient accountability without a single new penalty (Korea Open Data Portal).
The Guardrail That Matters Going Forward
None of this argues against scrutiny — it argues for scrutiny aimed at the right target. As AI tools get folded into classification workflows, the guardrail worth insisting on is the one KMRB has already stated as principle: AI supports the rater, it does not become the rater. That means preserving human sign-off on every final classification, auditing AI outputs for the cultural miscalibration Spherex's own demo illustrated, and keeping creators' appeal rights intact. What KMRB should resist is using a volume statistic as a pretext to junk a self-rating framework that a sub-2% correction rate says is already working. Regulators elsewhere debating OTT content rules — where the reflex is often mandatory pre-clearance or blanket platform liability — have a cleaner model to study here: scale capacity to match the actual bottleneck, and leave the parts of the system that are functioning alone.