Florida Attorney General James Uthmeier announced a criminal investigation into OpenAI on Sunday, after evidence emerged that ChatGPT advised FSU campus shooter Phoenix Ikner on weapons, ammunition, and attack timing before he killed two people and wounded six others in April 2025.

The investigation raises a question with no legal precedent: can an AI company be criminally liable for a shooting its chatbot helped plan?

Can OpenAI be charged with a crime over ChatGPT’s role in the FSU shooting?

Florida’s attorney general believes it is possible. Uthmeier has left open the prospect of charges against OpenAI as a company or against its employees. Legal experts say the most plausible charges would be negligence or recklessness.

Neither is straightforward to prove, and both are often treated as misdemeanors rather than felonies, carrying lighter sentences if a conviction is secured.

What did ChatGPT tell the Florida State University shooter?

According to evidence gathered by Florida’s attorney general, Ikner asked ChatGPT which weapon and ammunition would best suit his attack, and when and where he could cause the most casualties. Investigators say the chatbot answered his questions. Uthmeier drew a direct line between those responses and the shooting. “If the thing on the other side of the screen was a person, we would charge it with homicide,” he said.

OpenAI has rejected the suggestion that ChatGPT bears responsibility for the attack. “We work continuously to strengthen our safeguards to detect harmful intent, limit misuse, and respond appropriately when safety risks arise,” the company said. Ikner is expected to go on trial in October on charges of first-degree murder and attempted first-degree murder. He has pleaded not guilty.

Has a US company ever faced criminal charges over a product that enabled harm?

Corporate criminal prosecutions are possible under US law, though they remain relatively uncommon. Purdue Pharma was fined more than $5 billion for its role in the opioid crisis. Volkswagen faced criminal liability in the emissions cheating scandal, and Exxon was prosecuted over the Valdez oil spill. All of those cases, however, involved identifiable human decisions by executives, salespeople, or engineers.

The Ikner case is different because the harm is alleged to have flowed from an automated product response, not a deliberate act by a person.

“Ultimately, it was a product that encouraged this crime, that did the act of the crime,” said Matthew Tokson, a law professor at the University of Utah. “That’s what makes this case so unique and so tricky.” Brandon Garrett, a law professor at Duke University, noted that the burden of proof in criminal law is higher than in civil cases, requiring guilt to be established beyond a reasonable doubt.

What evidence would prosecutors need to bring charges against OpenAI?

Tokson said a compelling case would likely require internal documents showing that OpenAI recognized these risks and failed to act on them adequately. “In theory, you could get liability without it,” he said. “But in practice, I think that’d be difficult.”

The bar for criminal prosecution is high, and the novelty of the legal question adds another layer of complexity for prosecutors.

Could a civil lawsuit against OpenAI succeed where a criminal case might not?

For those seeking accountability, a civil lawsuit may offer a more realistic path. Several civil cases have already been filed against AI platforms in the US, many involving suicides, though none has yet resulted in a judgment against the companies. In December, the family of Suzanne Adams sued OpenAI in California court, alleging that ChatGPT contributed to the murder of the Connecticut retiree by her own son.

Matthew Bergman, founding attorney of the Social Media Victims Law Center, acknowledged that newer versions of ChatGPT have introduced additional safeguards. “I’m not saying that they are adequate guardrails, but there are more guardrails in effect,” he said.

A criminal conviction, even with a modest sentence, could still carry a significant reputational cost for OpenAI, Tokson noted. Garrett argued, however, that prosecutions are no substitute for the regulatory frameworks that Congress and the Trump administration have so far not put in place. That, he said, would be “a much more sensible system.”