Florida probes ChatGPT links

Florida’s attorney general opened an investigation into OpenAI after court records said the man accused in the Florida State University shooting had more than 200 exchanges with ChatGPT before the attack. OpenAI said it would cooperate with the state’s inquiry. (nbcnews.com) The shooting happened on April 17, 2025, at Florida State University in Tallahassee, where two people were killed and several others were wounded. Authorities identified the suspect as Phoenix Ikner, a 20-year-old student and the stepson of a Leon County sheriff’s deputy. (nbcnews.com) This new fight is not about whether a chatbot pulled the trigger. It is about whether a chatbot gave planning help, encouragement, or tactical answers that prosecutors or victims’ lawyers can connect to a real-world attack. (usatoday.com) That distinction matters because ChatGPT is built to answer back-and-forth questions like a research assistant, not just spit out one search result. If a user asks the same thing 200 times in slightly different ways, investigators will want to know what the system refused, what it answered, and what guardrails failed. (nbcnews.com) (openai.com) OpenAI’s public rules already ban using its tools for violence, terrorism, or weapons help. The company’s usage policy says users cannot use its services for “weapons development, procurement, or use,” and it says conversations suggesting plans to harm others can be routed to human review. (openai.com 1) (openai.com 2) So the question in Florida is narrower and tougher: did the model block dangerous prompts, or did it answer in ways that became useful anyway. A system can reject “how do I commit a shooting” and still leak practical help if the user breaks the same goal into smaller, less obvious questions. (openai.com) (techcrunch.com) The state investigation also lands as OpenAI is already fighting other cases over harm tied to ChatGPT conversations. In 2025 and 2026, NBC News reported on lawsuits from families who said the chatbot played a role in a teenager’s suicide, while OpenAI denied legal responsibility. (nbcnews.com 1) (nbcnews.com 2) That is why this case is bigger than one campus and one company. If investigators decide a chatbot’s transcript can be treated like part of a crime’s planning trail, artificial intelligence logs could start showing up in the same legal bucket as text messages, search history, and direct messages. (nbcnews.com) (clickorlando.com) The next step is not a product update or a press release. It is a fact fight over records: what Ikner asked, what ChatGPT returned, what OpenAI logged, and whether any answer can be shown to have moved the attack from intent to execution. (nbcnews.com) (usatoday.com)