Florida AG probes OpenAI

Florida’s top law-enforcement officer, James Uthmeier, said on April 9 that his office is opening an investigation into OpenAI and ChatGPT, and he said subpoenas are coming. He tied the probe to child-safety complaints, national-security fears, and records from the 2025 Florida State University shooting case. (politico.com) The shooting reference is specific, not rhetorical. Florida news outlets reported that court records in the Florida State University case showed accused gunman Phoenix Ikner had ChatGPT conversations that appeared to involve planning before the April 17, 2025 attack. (tallahassee.com) Uthmeier is not suing OpenAI yet. What he announced is an investigation, which is the stage where a state attorney general demands documents, internal policies, and answers before deciding whether to file a case or push for a settlement. (cbsnews.com) That matters because state attorneys general have become the people who test new rules before Congress does. When a technology moves faster than federal law, a state office can still use consumer-protection powers, subpoena power, and public pressure to force a company into the open. (axios.com) Florida’s argument has three parts. One part is that chatbots can expose minors to self-harm, sexual, or manipulative material; one part is that model data or tools could be exploited by foreign adversaries including China; and one part is that a product can be dangerous even if the company never intended the misuse. (orlandosentinel.com) OpenAI walked into this fight while trying to show the opposite image. On April 8, one day before Florida’s announcement, the company published a Child Safety Blueprint with the National Center for Missing and Exploited Children and said it wanted stronger reporting, detection, and built-in safeguards across the industry. (openai.com) So the timing is brutal for OpenAI. A company that spent Tuesday talking about child-protection standards was answering by Thursday to a state investigation that says those protections may not be enough in the real world. (openai.com, wfsu.org) There is also a money angle hanging over this. Reuters and CNBC reported that OpenAI is preparing for a potential initial public offering that could value the company at up to $1 trillion, which means any legal cloud now lands while investors are trying to price future risk. (reuters.com, cnbc.com) The bigger test is not whether Florida can prove a chatbot “caused” a crime. The harder question is whether states can force artificial-intelligence companies to treat misuse patterns the way carmakers treat crash data: as warnings that trigger redesigns, logs, and recalls of risky features before the next disaster. (techcrunch.com, politico.com)