Incident Overview
On April 2, 2026, during the live broadcast of NASA’s Artemis II launch, Korea’s public broadcaster, the Korean Broadcasting System (KBS), displayed AI-generated real-time translation subtitles that mistranslated critical aviation terms into Korean profanity. The terms ‘roger,’ ‘roll,’ and ‘pitch,’ which were part of mission control communications, were incorrectly interpreted by the AI system and rendered as vulgar language in the subtitles. This error occurred during a livestream on KBS’s YouTube channel, exposing viewers to inappropriate content in real time.
The incident was confirmed by KBS, which issued an official apology on April 2, acknowledging the mistake and attributing it to a phonetic similarity issue in the AI translation process. The broadcaster also announced plans to implement stronger profanity filters and temporarily paused live AI subtitling as a precautionary measure.
The harm from this incident included the broadcast of harmful and hateful content, which violated broadcasting standards and caused reputational damage to KBS. While no direct harm to individuals was reported, the exposure of profanity in a professional and scientific context—such as a space mission launch—highlighted the risks of unmoderated AI systems in live broadcasting environments.
The AI system implicated in the incident was a real-time translation subtitle generation system, deployed by KBS to provide live captions for international broadcasts. The system’s failure to accurately translate technical terminology underscored limitations in current AI natural language processing (NLP) capabilities, particularly in handling phonetically similar words or phrases.
This incident aligns with broader risks associated with AI system safety and failures, as categorized by frameworks such as the OECD AI Incidents and Hazards Monitor (AIM) and the Domain Taxonomy of AI Risks. Specifically, it falls under the category of ‘AI system safety, failures, and limitations,’ where unexpected outcomes arise from the deployment of AI in high-stakes or public-facing contexts.
Possible Root Cause
The primary root cause of the incident was identified by KBS as a phonetic similarity issue within the AI translation system. The terms ‘roger,’ ‘roll,’ and ‘pitch’—common in aviation communications—were misinterpreted by the AI due to their phonetic resemblance to Korean profanity. This suggests a failure in the system’s ability to distinguish between technically precise language and colloquial or offensive expressions, particularly in real-time processing environments.
The use of unmoderated live AI translation further exacerbated the issue. This lack of moderation allowed the erroneous subtitles to be displayed publicly without immediate correction.
Technical limitations in the AI’s NLP model may have contributed to the error.
Takeaways
The KBS incident serves as a case study in the risks of deploying AI systems in live broadcasting without adequate safeguards. It demonstrates how even minor phonetic or contextual errors can escalate into significant reputational and operational consequences, particularly when the content is disseminated to a wide audience in real time.
- Limits of Unmoderated AI in Live Environments: The incident underscores the dangers of relying solely on AI for real-time content generation without human oversight. Live broadcasts, by their nature, require immediate processing, which can amplify the impact of errors.
- Phonetic and Contextual Safeguards: AI translation systems should incorporate additional layers of validation, such as phonetic similarity checks or context-aware filters, to prevent unintended outputs.
- KBS’s prompt apology and announcement of corrective measures reflect a commitment to accountability, which can help mitigate reputational damage.
- Lessons for AI Developers and Broadcasters: Developers of AI translation systems should prioritize robustness in handling edge cases, particularly in specialized or technical language. Broadcasters, in turn, should conduct thorough pilot tests and stress simulations before deploying AI in live settings, especially for high-profile events like space missions or international broadcasts.
Sources
Incident summary from the AI Incident Database, used under CC BY-SA 4.0.