Florida prosecutor James Uthmeier announced on Tuesday the opening of a criminal investigation into OpenAI and ChatGPT. This includes shedding light on the role played by artificial intelligence in a deadly attack that took place in April 2025.

Published on: 04/21/2026 - 9:53 p.m. Modified on: 04/22/2026 - 09:42 a.m.

A first in criminal matters in the United States. Florida prosecutor James Uthmeier announced on Tuesday April 21 the opening of a criminal investigation into OpenAI and ChatGPT to shed light on the role of this AI in a fatal attack that occurred in April 2025.

According to information collected by the prosecutor's office, Phoenix Ikner, a student at Florida State University, exchanged with ChatGPT before shooting several people on campus, leaving two dead and six injured.

During a press conference, James Uthmeier explained that ChatGPT had "provided significant guidance to the shooter before he committed this hate crime." The conversational robot made suggestions on the right weapon and ammunition, as well as when and where to reach as many people as possible.

Also read: The calm before the storm Mythos, this serial AI detecting computer flaws

Based on this, "my investigators told me that if this thing on the other side of the screen (ChatGPT) was a person, we would charge him with homicide," he said.

ChatGPT “not responsible for this terrible crime”

Asked by AFP, OpenAI argued that “ChatGPT (was) not responsible for this awful crime” and that the interface had simply produced “factual responses” to the requests of the student whose trial is due to begin in October.

The group stressed that it spontaneously transmitted data concerning Phoenix Ikner after being informed of the attack.

OpenAI says it has been working for a long time to strengthen its safeguards to “detect dangerous intent” and “respond appropriately in the event of a risk to the safety” of people.

See also War in the Middle East: artificial intelligence at the heart of the conflict

The investigation announced Tuesday does not prejudge possible proceedings against OpenAI or certain of its members.

Since the arrival of ChatGPT in November 2022, quickly followed by competitors like Gemini (Google) or Claude (Anthropic), this new generation of interfaces has regularly been criticized for its supposed influence on certain users.

Parents, researchers and associations have particularly drawn attention to software presenting itself as virtual companions, in particular that of the start-ups Character.AI and Replika.

But even if ChatGPT, Claude or Gemini are more general products which are not intended to play a role, they are used as such by many adolescents and adults.

“It is not because it is a chatbot that there should not be criminal liability”

Prosecutions were brought before the civil courts, in most cases for suicides and not homicides. In January, Google and Character.AI reached out-of-court settlements with families accusing chatbots of harming minors and causing one of them to end his life.

In March, the family of a young girl seriously injured during the Tumbler Ridge massacre (western Canada) announced that they were taking civil action against OpenAI for negligence. She criticizes the company for not having reported to the police worrying messages written on ChatGPT by the alleged author of the attack.

OpenAI said it has implemented a new security protocol that would trigger an alert based on a similar conversation.

Also read: Anthropic against the Trump administration, a decisive standoff over the use of AI by the army

The indictment of a legal entity is provided for by American law, but prosecutor Uthmeier, appointed by Republican Governor Ron DeSantis, recognized “that with AI, we are venturing into uncharted territory”.

The representative of the public prosecutor sent OpenAI a request for documents, in particular concerning its internal regulations and its methods of supervising AI models. “It is not because it is a chatbot that there should not be criminal liability,” insisted the prosecutor, specifying that civil proceedings would continue their course.

According to James Uthmeier, the exchanges analyzed by his teams showed that the suspect “(had) described his plan” of attack to ChatGPT. He notably asked the interface how to arouse the interest of the national media in his gesture, and “ChatGPT offered him an analysis,” said the prosecutor.

“There is a difference,” he said, “between an internet provider or a mobile equipment manufacturer and an app that produces what looks like practical advice for a school shooting.”

With AFP

Keywords associated with the article

The content you are trying to access does not exist or is no longer available.