ChatGPT Reportedly Gave Incorrect Mount Misen Ropeway Hours, Leaving Couple Stranded After Hike in Japan

Dana Yao and her husband, Robert, reportedly used ChatGPT to plan a sunset hike on Mount Misen in Japan. ChatGPT allegedly told them the last ropeway departed at 5:30 PM, but the service had already closed when they reached the summit, leaving them stranded.

Risk classification

• Primary risk domain: 3 Misinformation
• Primary risk subdomain: 3.1 False or misleading information

The AI systems generated false and misleading travel information, leading users to form inaccurate beliefs that resulted in material and physical risks.

Additional risk subdomains

• 5.1 Overreliance and unsafe use: Travelers relied on general-purpose AI chatbots for safety-critical planning, such as mountain hiking schedules and high-altitude routes.
• 7.3 Lack of capability or robustness: The underlying failure stems from the AI models' lack of capability to distinguish fact from statistical word association, leading to hallucinations.

Causal factors

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

The incidents were caused by deployed AI systems generating unintended hallucinations and false information during user interactions.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The systems described are general-purpose chatbots and generative AI tools used to interact with humans and generate content, which fall under transparency obligations.

AI system and alleged parties

• AI system: ChatGPT, Gemini, Layla (Google, OpenAI)
• AI purpose: Navigation Assistant; Chatbot
• Behaviour type: Assistant
• Alleged developer: OpenAI, Large language model developers, Chatbot developers
• Alleged deployer: Travelers, tourists, OpenAI, Dana Yao's husband Robert, Dana Yao, ChatGPT users, Chatbot users
• Alleged harmed parties: Travelers, tourists, Epistemic integrity, Dana Yao's husband Robert, Dana Yao, ChatGPT users, Chatbot users

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 Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Minor
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Epistemic

Reported: The reports explicitly describe epistemic harm through the fabrication of locations, schedules, and routes by AI systems.

Directly caused: AI systems directly fabricated non-existent locations like the 'Sacred Canyon of Humantay' and a fake cable car in Malaysia, and provided false operating hours.

Indirectly caused: The proliferation of these hallucinations blurs the lines between reality and synthetic information, eroding trust in digital travel information.

Inferred additional harm: Widespread use of generative AI for search and planning is likely polluting the broader information ecosystem with fabricated facts, affecting millions of users.

People affected

• Occurrences reported: 1
• People reportedly harmed: 8
• People reportedly exposed: 9

Potential causes

Management

• Prioritizing Engagement Over Safety: Platforms use AI to edit media and boost engagement without safety checks.
• Release of Unsafe Conversational AI: Tech firms deploy LLMs for travel planning despite known hallucination issues.

Technology

• Inherent LLM Hallucinations: LLMs statistically predict words and fabricate non-existent places.
• Lack of Physical World Understanding: AI cannot distinguish between city walks and dangerous mountain climbs.
• Identical Presentation of Facts: AI presents fabricated errors and real facts with the same confident tone.

Data Inputs

• Outdated Travel Information: AI used outdated data, leading to incorrect ropeway and restaurant details.
• Unverified AI-Generated Media: Social media videos generated by AI fabricated non-existent tourist sites.
• Statistical Word Association: Combining regional words to create fictional destinations like Humantay Canyon.

Human Factors

• Overreliance on AI Output: Travellers trusted the AI itinerary without verifying safety or feasibility.
• Lack of Local Destination Knowledge: Travellers planned trips to unfamiliar areas making verification difficult.
• Misleading Confident Tone: Users were swayed by the vivid adjectives and convincing writing of the AI.

Process and Methods

• Inadequate Manual Verification: Planning relied solely on AI without double-checking official sources.
• Lack of Mitigation Strategies: Travellers failed to prepare backup plans for when AI information failed.
• Inefficient Search Methods: Verifying AI output manually was as laborious as traditional planning.

Regulatory Environment

• Lack of Standardized Watermarking: Absence of clear indicators to distinguish AI-generated travel content.
• Slow Regulatory Implementation: Proposed US/EU regulations on AI content identification are not yet active.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide clear, consistent, and detailed anecdotal evidence of AI travel planning failures across multiple platforms (ChatGPT, Layla) and geographic locations (Peru, Japan, Italy). The mechanisms of failure (LLM hallucinations) are well-explained by experts quoted in the text.
• Ambiguities identified: None of significance; the role of the AI in generating the false information is explicitly stated.
• Alternative interpretations: None.

The incident involves consumer-grade generative AI tools producing hallucinated travel information, leading to minor financial losses and localized safety risks for individual tourists. There are no national security implications, as the event does not threaten critical infrastructure, sovereignty, intelligence security, economic stability, or societal cohesion.

• Overall national security impact: Negligible
• Scope: Multiple nations
• Primary target: No clear primary
• Other affected: Peru, Japan, Italy, Malaysia
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. This is an ongoing, persistent characteristic of LLM technology rather than an active national security crisis requiring urgent response.
• Autonomy: Human-controlled. The AI acts as an informational assistant, and humans retain full control over whether to execute or verify the suggested itineraries.
• Novelty: Established threat. AI hallucinations and generation of false information are well-established characteristics of large language models since their widespread deployment.

Impact by dimension

• Physical security: Negligible. Minor localized safety incidents involving individual tourists, with no impact on critical infrastructure or national physical security.
• Information security: Negligible. Hallucinations are unintentional, non-malicious errors in consumer travel tools, not coordinated information warfare or intelligence compromise.
• Sovereignty: Negligible. No impact on state authority, electoral systems, border control, or core government decision-making processes.
• Economic security: Negligible. Financial losses are limited to minor individual travel expenses and do not threaten national economic stability or strategic industries.
• Societal stability: Negligible. While 33 percent of users reported false recommendations, this represents consumer inconvenience rather than systematic human rights violations or threats to societal stability.