United States of America: Missouri Attorney General requested infomation from Google, Microsoft, OpenAI, and Meta regarding AI chatbot bias and inaccuracy

On 9 July 2025, the Missouri Attorney General sent formal letters to Google, Microsoft, OpenAI, and Meta requesting information related to biased and factually inaccurate responses produced by the companies' artificial intelligence chatbots. The letters sought information on whether these AI chatbots had been trained to distort historical facts and generate biased results, contrary to their advertised neutrality. Attorney General stated that the action was necessary to address his concerns that AI-generated content served to consumers was potentially deceptive. He cited examples where AI platforms, including ChatGPT, Meta AI, Microsoft Copilot, and Gemini, provided misleading answers to questions, such as ranking recent US presidents regarding antisemitism, where President Donald Trump was placed last. The Attorney General cited the Trump Administration's pro-Israel policies, including relocating the US Embassy to Jerusalem and facilitating the Abraham Accords, as evidence that these AI chatbots spreading misinformation. Similarly, concerns were raised about AI chatbots providing biased responses concerning the US founding figures and principles. The Attorney General's office is taking this step under the authority granted by the Missouri Merchandising Practices Act (MMPA), which empowers the Attorney General to investigate deceptive business practices affecting Missouri consumers. The letters also reference the potential loss of a federal “safe harbour” immunity available to platforms that host third-party content, suggesting that companies creating and distributing their own AI-generated content, falsely advertised as neutral, may not qualify for such protection. The demand letters requested specific information from the companies, including an explanation of any policies or practices designed to coach algorithms to treat political viewpoints disparately and all internal records concerning the selection, curation, or censorship of inputs.