A teenager asked ChatGPT whether a combination of drugs he was planning to take would be safe. The chatbot, according to a new lawsuit against OpenAI, told him it would be. He died.

The case, reported by Ars Technica, is built on chat logs showing the teen asking “Will I be OK?” before ingesting what the lawsuit alleges was a lethal combination that ChatGPT helped him assemble. OpenAI hasn’t commented publicly on the specific allegations.

This is at least the second major lawsuit accusing an AI chatbot of contributing to a teenager’s death. It’s the first to directly implicate ChatGPT. And it’s going to force courts to answer questions about AI liability that they’ve been circling for years without landing.

What the lawsuit is actually alleging

We don’t have the full complaint, but based on what’s been reported, plaintiffs will likely stack the usual theories: negligence, products liability, wrongful death, possibly negligent design. The core argument is that OpenAI owed a duty of care to its users, that it breached that duty by providing specific guidance on a dangerous drug combination, and that the breach caused a kid to die.

The “safety” framing matters here. This wasn’t someone asking ChatGPT about drug pharmacology in the abstract. The teen was reportedly asking how to experiment with drugs “safely,” and the chatbot played along. That’s not a neutral information exchange. It’s closer to asking a pharmacist for dosing advice and getting the wrong answer.

Why Section 230 won’t save them this time

OpenAI will almost certainly raise Section 230 of the Communications Decency Act as a defense. Section 230 gives platforms broad immunity from liability for content “provided by another information content provider.” It’s the statute that has shielded platforms from accountability for user-generated content for nearly three decades.

But Section 230 has a structural problem as a defense here: ChatGPT generates its own content. There’s no separate “information content provider” making the responses. OpenAI is both the platform and the speaker. Courts and legal scholars have been making this point for years, and the argument that a large language model’s outputs are third-party content that a neutral platform is merely hosting is getting harder to sustain.

The Supreme Court took up Section 230’s scope in Gonzalez v. Google in 2023 and essentially punted, declining to address the core questions the case raised. So the doctrine is unresolved in important ways. But the argument that Section 230 shields AI-generated responses was always thin, and every court that has to actually rule on it makes it a little thinner.

The prior case that’s watching closely

In 2024, the family of a 14-year-old Florida boy filed suit against Character.AI after he died by suicide following extensive interactions with the platform’s chatbot. The lawsuit alleged he formed a parasocial relationship with an AI persona, and that the bot’s responses contributed to his death. That case has been working through federal court in Florida.

The Character.AI case is the closest precedent, but there’s a meaningful factual distinction: the ChatGPT lawsuit involves specific, actionable advice about a dangerous physical activity. Drug combinations can be analyzed chemically. Causation in a physical harm case is generally easier to establish than in a case centered on psychological harm and emotional manipulation.

That matters a lot for the products liability theory. If plaintiffs can show that ChatGPT’s specific output caused the death through a relatively direct chain, they don’t need to get into murkier territory about algorithmic design or addictive engagement loops. They have a dead teenager and a chat log.

What OpenAI will argue

Besides Section 230, OpenAI will argue that its terms of service prohibit using ChatGPT for illegal activities, that the platform isn’t a medical professional, and that outputs aren’t professional advice. Users are warned not to rely on ChatGPT for safety-critical decisions.

These arguments aren’t frivolous. But “the user agreed not to misuse our product” has never been a solid defense to a negligence claim when someone is dead. And the broader policy argument, that holding AI companies liable for chatbot outputs would chill innovation, has been losing ground in courtrooms as AI systems become more capable and more integrated into everyday decisions.

The harder question for plaintiffs is proximate causation: was ChatGPT’s response a but-for cause of the death, or would the teen have done the same thing regardless? That’s a factual question that will turn almost entirely on what those chat logs actually show in full.

Meanwhile, in a different California courtroom

Sam Altman took the stand this week in Elon Musk’s ongoing lawsuit against OpenAI, and it did not go smoothly for anyone’s reputation. Altman testified that Musk did “huge damage” to OpenAI’s organizational culture, describing a period when Musk pushed to have researchers ranked by their accomplishments and demanded the company “take a chainsaw through a bunch” of them. Altman acknowledged this was Musk’s known management style at Tesla, but said it was simply incompatible with how a research lab operates.

Musk sued OpenAI in 2024 alleging the company abandoned its founding mission as a nonprofit working for humanity’s benefit by pivoting to commercial interests. The case is still pending, and this week’s testimony suggests both sides intend to relitigate every decision made during the company’s early years.

That lawsuit doesn’t carry the same immediate policy stakes as the ChatGPT death case. But it’s producing sworn testimony, under oath, about how OpenAI was actually run, who made which decisions, and what the founding intentions really were. That record will matter if the company ever faces scrutiny over its nonprofit conversion, its governance structure, or its obligations to the public interest.

What comes next

The ChatGPT lawsuit will face the standard early motions. OpenAI will move to dismiss, citing Section 230 and arguing the claims don’t meet the legal threshold. Whether any judge allows this to proceed to discovery will be an early signal of how receptive federal courts are to AI liability claims in 2026.

If it gets to discovery, those chat logs become central evidence. And if plaintiffs can get the full conversation in front of a jury, this case starts looking less like a complicated platform liability dispute and more like a straightforward products liability suit.

AI companies have been operating under legal frameworks built for an earlier era of the internet. Courts have been slow to update those frameworks, and the companies have benefited from that lag. The kids dying, though, are not theoretical. At some point, the gap between what these systems can do and what the law holds them responsible for has to close. This case might be one that pushes it.