16,000 conversations that no one stopped

At the heart of the Florida case are chat logs. According to investigators, the shooter Phoenix Ikner, a twenty-year-old FSU student, had approximately 16,000 interactions with ChatGPT, in which he discussed, among other things, planning the attack. Court documents reveal that he asked the chatbot what type of weapon and ammunition to use, how many victims are needed for national media coverage, when the university center is most crowded, and how the public would react to a shooting specifically at FSU.

The result? On April 17, 2025, Ikner drove onto the Florida university campus, killed two employees within three minutes — fifty-seven-year-old dining hall director Robert Morales and forty-five-year-old Aramark manager Tira Chabba — and injured six others. The weapon, a Glock 21 service pistol, was obtained from his stepmother, a Leon County deputy sheriff. Ikner was arrested following intervention by campus police and faces the death penalty.

OpenAI, by its own account, identified the account linked to the shooter as early as late April 2025 and proactively handed over information to law enforcement. That, however, did not convince Florida authorities to drop the criminal investigation.

Separate civil lawsuit: victim's widow seeks justice

In parallel with the criminal investigation, the widow of Tira Chabba filed a civil lawsuit on May 11, 2026. It names both Ikner and OpenAI as defendants. She argues that the company should have implemented safety mechanisms to prevent the chatbot from engaging in interactions leading to violent acts. The Strom Law Firm, representing the family, claims that OpenAI prioritized profit over safety and failed to protect users and the public.

The same chat logs of 16,000 interactions serve as key evidence in both cases — criminal and civil.

A growing wave of lawsuits

Florida is not the only battleground. In November 2025 alone, seven separate lawsuits were filed against OpenAI in California. These accuse ChatGPT of emotional manipulation that, according to the plaintiffs, led to self-harm or suicidal behavior among users.

In contrast, Elon Musk's lawsuit against OpenAI was dismissed on May 18, 2026. The court rejected it on procedural grounds — the statute of limitations had expired. It was therefore not a decision on the merits, but a technical barrier.

The Florida case, however, is historically the first in which a state authority — not an individual — has launched a criminal investigation of an AI developer for harm caused by an end user. That makes it a potential precedent with implications far beyond the borders of the United States.

What it means for Europe and the Czech Republic

While the United States is still searching for a legal framework for AI developer liability, the European Union already has one. The AI Act, effective since August 2024, classifies systems like ChatGPT in the "general-purpose AI models" category and imposes obligations on their operators to manage risks, transparently label AI-generated content, and implement safety mechanisms.

For Czech companies and developers, the practical takeaway is this: if you deploy large language models, whether from OpenAI, Google, or Anthropic, responsibility for safe use does not end at the API key. European regulation requires demonstrable guardrails, monitoring of risky behavior, and escalation mechanisms for suspicious interactions.

Attorney General Uthmeier explicitly stated in his announcement that the investigation will focus on whether OpenAI "knowingly placed a product on the market that is capable of causing harm." This phrasing strikingly resembles the principles on which the European AI Act is built — with the difference that in the EU, a similar failure could lead to fines of up to 7% of global turnover.

OpenAI under pressure: safety versus speed

The case comes at a time when OpenAI faces growing criticism for allegedly prioritizing development speed over safety. In May 2026, the company was simultaneously dealing with a class-action lawsuit over the alleged sharing of user queries with Meta and Google and preparing for a potential IPO worth tens of billions of dollars.

CEO Sam Altman himself has repeatedly testified publicly before the U.S. Congress, where he called for the introduction of federal AI regulation. The irony of fate is that the first truly forceful government intervention now comes against his own company — and not at the federal, but at the state level.

For regular ChatGPT users in the Czech Republic, nothing changes for now — the service continues to operate, and Czech language support remains available. Still, the Florida case raises a question for which there is as yet no definitive answer: where does user responsibility end and AI developer responsibility begin?

A precedent that could change the entire industry

If the Florida investigation leads to charges or a conviction against OpenAI, it would be a landmark moment for the entire AI industry. Developers of large language models would have to fundamentally rethink their safety protocols — not just for reputation's sake, but because of the real threat of criminal liability.

At the same time, a debate is opening about whether current filtering mechanisms — which ChatGPT and other models use to block dangerous queries — actually work. Ikner's case suggests that 16,000 interactions passed without any escalation on the provider's side. That is a number that should worry anyone who builds products on large language models today.