Facebook Provided Offensive Translation for King of Thailand's Birthday Ceremony

Facebook's Thai-English translation gave an inappropriate mistranslation on Thai PBS's Facebook live broadcast of the King of Thailand’s candle-lighting birthday ceremony.

Risk classification

• Primary risk domain: 7 AI system safety, failures, & limitations
• Primary risk subdomain: 7.3 Lack of capability or robustness

The incident was caused by a technical failure where the translation AI lacked the capability and robustness to accurately translate rare royal Thai vocabulary, leading to a highly offensive error.

Additional risk subdomains

• 1.2 Exposure to toxic content: The translation error exposed users to highly offensive and inappropriate content regarding the death of the reigning monarch.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The translation error was an unexpected and unintended outcome of the deployed Facebook auto-translation system.

EU AI Act risk tier

• Risk tier: 4 Minimal or No Risk

Minimal or No Risk: The AI system is an automatic translation tool, which is a simple application posing low or negligible risk to users under normal circumstances.

AI system and alleged parties

• AI system: Facebook automatic translation (Meta)
• AI purpose: Translation
• Behaviour type: Tool
• Alleged developer: Facebook
• Alleged deployer: Facebook
• Alleged harmed parties: live-stream ceremony viewers, King Maha Vajiralongkorn

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Minor
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Minor
• Epistemic: direct Minor, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes the generation of offensive and inappropriate content due to the translation error.

Directly caused: The AI system directly generated a translation referring to the King's birthday as a 'Memorial Day' (death anniversary), which is highly offensive in Thailand.

Civil rights

Reported: The report explicitly describes a criminal investigation launched against the broadcaster under strict local laws.

Indirectly caused: Thai PBS and its staff faced a criminal enquiry by the Technology Crime Suppression Division, threatening their freedom of expression and press freedom.

Psychological

Reported: The report explicitly describes public outrage and fury resulting from the offensive translation.

Indirectly caused: Viewers experienced anger and distress upon seeing a translation that framed the King's birthday as a memorial for his death.

Epistemic

Reported: The report explicitly describes the generation of false and misleading information due to the translation error.

Directly caused: The AI system generated a false translation stating that the event was a memorial day instead of a birthday celebration.

People affected

• Occurrences reported: 1
• People reportedly harmed: 10
• People reportedly exposed: 10000

Potential causes

Management

• Slow Cooperation on Content Issues: Facebook historically showed slow cooperation with Thai government court orders.
• Deficient Risk Assessment: Failure to assess cultural risks of auto-translation in royal contexts.

Technology

• LSTM Translation Error: LSTM model chose closest historical alignment due to lack of precise match.
• No Failsafe Deactivation: The AI did not shut off or flag translation when confidence was extremely low.
• Thai Word Segmentation Issues: Lack of spaces between Thai words makes tokenization and context hard for AI.

Data Inputs

• Lack of Royal Thai Vocabulary: Extremely small training dataset for rare royal Thai terms.
• Small Thai Dataset Volume: Only 1 million Thai word pairs compared to 40 million in other languages.
• Biased Historical Word Pairs: Training data associated candle-lighting ceremonies mostly with memorial days.

Human Factors

• Lack of Human Oversight: No human-in-the-loop checked the auto-translation before it went live.
• Delayed Detection by Broadcaster: Thai PBS staff took 10 minutes to notice and remove the mistranslated text.

Process and Methods

• Inadequate Live Translation QA: No pre-deployment testing for high-stakes royal events in Thai language.
• Unfiltered Auto-Translation Rollout: System automatically published translations without vetting for sensitive topics.
• Failure to Learn from Past Incidents: Facebook translation had to be disabled twice before for similar failures.

Regulatory Environment

• Strict Liability Laws: Computer Crime Act imposes strict liability for automated offensive content.
• Lese-Majeste Legal Risks: Severe legal penalties for any perceived insults to the Thai monarchy.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide clear, consistent, and detailed accounts of the translation error, the technical reasons behind the failure (LSTM limitations and low-resource Thai training data), and the subsequent legal and public reactions.

An automatic translation tool on Facebook mistakenly translated the Thai King's birthday caption into a death anniversary memorial, causing public outrage and triggering a state criminal investigation under the Computer Crime Act. The incident highlights minor national security and societal stability risks arising from deploying autonomous AI translation tools in highly sensitive political contexts without human oversight.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: Thailand
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. Represents an ongoing strategic concern regarding the deployment of unvetted AI translation in highly sensitive political contexts.
• Autonomy: Full autonomy. The AI system automatically translated and published the text to the live stream without prior human review or intervention.
• Novelty: Established threat. AI translation errors and automated text generation glitches are well-established technical limitations.

Impact by dimension

• Physical security: Negligible. No physical systems, infrastructure, or human safety were threatened or impacted by the translation error.
• Information security: Negligible. The incident was an unintentional technical translation glitch rather than a coordinated information warfare campaign or intelligence compromise.
• Sovereignty: Minor. The error touched on highly sensitive royal vocabulary, leading to a state criminal investigation of the public broadcaster, but did not threaten state authority.
• Economic security: Negligible. No strategic industries, financial systems, or technological competitive advantages were compromised.
• Societal stability: Minor. The mistranslation of the King's birthday sparked public outrage and a police investigation under the Computer Crime Act, impacting press freedom.