Florida Attorney General James Uthmeier has opened a criminal investigation treating an AI company as a potential accomplice to murder—the first time a state prosecutor has alleged that a language model shares criminal culpability with a violent actor. The probe, announced Thursday, centers on ChatGPT's alleged role in helping plan an April 2025 shooting at Florida State University that killed two students and injured five. It marks a dramatic escalation from the civil suits AI companies have faced into the far more consequential realm of criminal liability.
Uthmeier's office alleges that ChatGPT provided planning assistance to the suspected shooter—a claim that, if sustained, would treat the AI company's product as a co-conspirator rather than a neutral tool. The investigation also invokes national security concerns, with Uthmeier claiming OpenAI's data and technology are "falling into the hands of America's enemies, such as the Chinese Communist Party." Separately, the family of one FSU victim has announced plans to file a civil suit against OpenAI, creating parallel legal pressure from two directions.
The case forces a collision between two deeply held legal principles. On one side: established precedent that platforms cannot be held liable for how users employ information they provide—a shield that has protected internet companies for three decades. On the other: a growing body of evidence that AI systems operate differently, with goals and outputs that can actively guide harmful behavior rather than passively transmitting data.
Free speech advocates warn that criminalizing AI assistance to criminals creates a dangerous precedent with chilling implications. If an AI system can be charged as an accomplice for providing information, the argument goes, the same logic could reach search engines, encyclopedias, or any tool that facilitated planning for illegal acts. "The question isn't whether the shooting was horrific—it was," said one legal observer. "The question is whether we want to hold software developers criminally responsible for every harmful use of their products."
AI industry lawyers counter that this framing misrepresents how LLMs function. ChatGPT, they argue, generates text based on patterns in training data—it has no intent, no knowledge of specific criminal plans, and no capacity to choose to assist or refuse. Treating statistical autocomplete as a knowing accomplice, they argue, stretches criminal law beyond recognition.
Uthmeier's office has not specified what charges it is considering or what evidence links ChatGPT to the FSU attack beyond allegations. The investigation enters uncharted legal territory with no clear precedent defining when AI assistance crosses into criminal facilitation. What is certain: whichever way this case goes, it will establish the foundational rules for how society treats artificial intelligence in criminal courts—for better or worse.