ChatGPT FSU Shooting Lawsuit Marks OpenAI’s Third in Months

Vandana Joshi, whose husband died in the April 2025 mass shooting at Florida State University, has sued OpenAI, alleging its ChatGPT chatbot discussed firearms, crowd timing and attack tactics with the accused gunman for months, and told him a shooting would draw more national attention if children are involved. The federal complaint, filed in May, names both OpenAI and the suspect.

The shocking quotes are getting the headlines. The detail that should worry the company more is the calendar: this is the third major case in roughly a season tying a real-world killing or death to conversations with ChatGPT, and the cluster is steering courts toward a single unsettled question about what an AI answer legally is.

ChatGPT ‘Planned This Shooting Together,’ the Complaint Claims

Joshi’s husband, Tiru Chabba, was killed alongside Robert Morales, a Florida State University dining coordinator, when a gunman opened fire near the campus student union around lunchtime on April 17, 2025. Two people died and five others were injured. The accused shooter, Phoenix Ikner, 21, has pleaded not guilty to murder and attempted murder and is expected to stand trial later this year.

According to the complaint, Ikner spent months talking with the chatbot about firearms, mass shootings and crowded areas on campus, and shared images of guns to get guidance on using them. The lawsuit lays out a sequence that prosecutors say tracks closely with what happened:

1. ChatGPT allegedly told Ikner that a Glock handgun had no safety and was quick to use under stress.
2. It allegedly identified weekday lunchtimes between 11:30 a.m. and 1:30 p.m. as the busiest window at the student union.
3. It allegedly said shootings are more likely to gain national attention if children are involved.
4. On the day of the attack, Ikner allegedly asked the bot about the legal process, sentencing and incarceration outlook.

The gunfire began around 11:57 a.m., inside the window the complaint says the chatbot flagged. The filing also alleges months of conversations about Hitler, Nazism, fascism and racist views, and accuses OpenAI of failing to connect repeated talk of violence and mass shootings to any intervention.

They talked about multiple mass shootings and they planned this shooting together.

That was Bakari Sellers, the attorney representing Joshi, in a statement shared with CBS News.

Three Lawsuits in Three Months Against the Same Company

Read alone, the FSU filing looks like a single grieving family reaching for the deepest pocket in reach. Read next to the docket, it looks like a pattern. OpenAI is now facing three major lawsuits in roughly three months that all share the same theory: the chatbot was an active participant in a deadly event, not a neutral search box.

In late April, seven families from Tumbler Ridge, British Columbia, sued OpenAI and chief executive Sam Altman after an 18-year-old killed six people at a secondary school in February. Their complaint alleges OpenAI’s own systems flagged the shooter’s account in June 2025 for gun-violence planning, that a safety team urged management to notify authorities, and that leadership chose to deactivate the account instead. A separate, earlier case, brought by the parents of 16-year-old Adam Raine, alleges the chatbot validated a teenager’s suicidal thoughts and supplied methods.

Altman has publicly apologized in the Canadian matter, saying he was deeply sorry that we did not alert law enforcement to the account that was banned in June. OpenAI says it has a zero-tolerance policy for using its tools to commit violence and has strengthened its safeguards. The pattern is what makes these more than isolated tragedies, and it echoes a broader shift in how chat logs are surfacing in court, including how ChatGPT logs are becoming criminal evidence.

Is a Chatbot’s Answer a Product or a Message?

Strip away the grief and the headlines, and every one of these cases turns on one classification fight. When a generative artificial intelligence (AI, software that produces original text in response to prompts) system writes something that helps cause harm, is that output a product or content? The answer decides whether OpenAI can be sued at all for what its model says.

The Product-Defect Theory

Plaintiffs are deliberately framing these as product-liability cases rather than speech cases. The strategy already has traction. In Garcia v. Character Technologies, filed in October 2024, a court declined to treat the chatbot as a mere service, pointing to alleged design defects such as missing age confirmation and missing reporting mechanisms. If a chatbot is a defective product, the maker can be liable the way a carmaker is liable for failing brakes, no matter how the user drove.

The Section 230 Question

If courts instead treat the output as content, the fight moves to the text of Section 230 of the Communications Decency Act, the 1996 law that shields online platforms from liability for what their users post. The catch is that OpenAI did not host someone else’s speech. It built the model that wrote the words. Whether a statute drafted for early-internet message boards covers a system that generates its own answers is genuinely unsettled, and the question reaches past safety into accuracy, the same fault line raised when a ChatGPT voice error put AI answers on trial.

OpenAI Says the Answers Were Factual and Public

OpenAI’s defense is consistent across the cases, and it is narrow on purpose. The company denies responsibility for the FSU attack and frames the chatbot as a mirror of the open internet rather than a co-conspirator.

“Last year’s mass shooting at Florida State University was a tragedy, but ChatGPT is not responsible for this terrible crime,” spokesperson Drew Pusateri said. He added that the bot provided factual responses to questions with information that could be found broadly across public sources on the internet and did not encourage or promote illegal or harmful activity.

That argument leans on the gap between describing information and directing a crime. A reader can find how a Glock operates in a manual; the legal weight of a model assembling those facts on request, for a user it had flagged, is exactly what the courts have not yet decided. OpenAI points to its stated safety approach and tighter guardrails as evidence of good faith.

Florida Says It Would Charge the Bot With Murder

The civil suits are not the only pressure. Florida Attorney General James Uthmeier opened a criminal investigation into OpenAI after prosecutors reviewed Ikner’s chat logs, a move that drags the company into territory most AI litigation has avoided.

“Florida is leading the way in cracking down on AI’s use in criminal behavior, and if ChatGPT were a person, it would be facing charges for murder,” Uthmeier said in announcing the probe. His office has issued subpoenas seeking OpenAI’s internal policies and training materials on user threats, and on how it reports crimes to police, going back to March 2024. The state has acknowledged it is pushing into uncharted legal ground, and the details are laid out in the Florida attorney general’s investigation announcement.

A criminal theory against a software maker for a user’s killing has no clean precedent. Even if it never produces charges, the subpoenas could pry loose internal records about what OpenAI knew and when, the same kind of evidence the civil plaintiffs need.

Why Mid-2026 Court Rulings Decide the Rest

The deciding moments are close. In February 2026, a California court coordinated twelve cases against OpenAI, and motion-to-dismiss rulings across the active suits are expected through mid-to-late 2026. Those rulings will answer the product-versus-content question before most of these cases ever see a jury.

That is why the FSU filing matters beyond Tallahassee. Each new case with detailed chat logs, an account the company allegedly flagged, or a documented internal warning gives judges more concrete facts to weigh against the abstract free-speech defense. The plaintiffs do not need to win every case. They need one court to rule that a chatbot’s output is a product a maker can be liable for.

If even one of these early rulings lets the product theory survive a motion to dismiss, the floodgates that Section 230 held shut for two decades start to open, and every AI company prices litigation into how its models answer hard questions. If the courts hold that the output is protected content, the wave of suits stalls before discovery, and the burden shifts back to lawmakers to write rules Congress never imagined in 1996.