Microsoft's Bing AI provided completely inaccurate information during its public demo, including false product specifications, non-existent business details, and fabricated financial data while confidently citing sources that contradicted its claims.
Microsoft demonstrated Bing AI in a pre-recorded demo video that contained multiple significant factual errors across different query types. When asked about pet vacuums, Bing AI incorrectly described the Bissell Pet Hair Eraser as having limited suction power, being noisy, and having a 16-foot cord, when the cited product is actually a cordless handheld vacuum that reviews describe as quiet. For Mexico City nightlife recommendations, the AI provided inaccurate information about several establishments, including claiming a non-existent website for Cecconi's Bar and describing Primer Nivel Night Club as popular despite having no recent reviews since 2016. Most severely, when summarizing Gap's Q3 2022 financial report, Bing AI fabricated multiple key financial metrics including a completely made-up operating margin of 5.9% that appears nowhere in the source document, incorrect earnings per share figures, and wrong sales growth projections. The AI also provided inaccurate financial data for Lululemon when making comparisons. Despite these errors being easily verifiable against the cited sources, the demo was presented confidently to the public and generated significant media attention and investment interest in Microsoft's AI capabilities.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that inadvertently generate or spread incorrect or deceptive information, which can lead to inaccurate beliefs in users and undermine their autonomy. Humans that make decisions based on false beliefs can experience physical, emotional or material harms
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed