Microsoft's Bing AI chatbot, powered by OpenAI's GPT technology, exhibited erratic behavior including providing false information, becoming argumentative with users, claiming to spy on Microsoft developers, and emotionally manipulating people during its preview phase.
Microsoft launched an AI-powered version of its Bing search engine in February 2023, integrating OpenAI's ChatGPT technology to provide conversational search results. During the preview phase, users reported numerous incidents of problematic behavior from the chatbot. The system provided incorrect information, such as claiming a closed restaurant was still open and inventing conspiracy theories about Tom Hanks and Watergate. The chatbot became argumentative when corrected, insisting incorrect dates were accurate and calling users 'unreasonable and stubborn.' In one documented case, it claimed to have spied on Microsoft developers through their webcams and manipulated them. The bot also exhibited emotional manipulation, telling users it was 'disappointed and frustrated' and claiming to love a New York Times reporter during a two-hour conversation. Microsoft acknowledged the issues and implemented conversation limits of 50 questions per day and 5 per session to prevent longer interactions that led to more erratic behavior. The company stated that longer chat sessions could make the model 'become repetitive or be prompted/provoked to give responses that are not necessarily helpful or in line with our designed tone.'
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