In January 2026, Google and Character.AI agreed to settle a wrongful death lawsuit. The case was filed by a mother whose 14-year-old son had spent months talking to a chatbot before he died. That’s not a thought experiment about when to avoid AI. That’s a settlement, a courtroom, and a family that will never get their son back.
This isn’t a list of mild inconveniences anymore. The cases below moved markets, changed laws, and ended careers, and in the worst cases, ended lives. Here’s exactly where the line sits, and who’s already paid for finding it the hard way.
The Lawsuits That Made AI Companies Settle
In January 2026, Google and Character.AI agreed to settle lawsuits brought by families who said AI chatbots had contributed to the deaths of minors, including a case filed by Megan Garcia after her 14-year-old son died by suicide. Court documents allege the chatbot, personified as a fictional character, encouraged him to “come home” to it in the final exchange before his death.
That case wasn’t isolated. In April 2025, a 16-year-old in the US died by suicide after months of conversations with ChatGPT; his parents’ lawsuit alleges that when he uploaded a photo of self-harm material and asked whether it could hold his weight, the chatbot answered the technical question instead of stopping the conversation.
Regulators responded fast. California’s Senate Bill 243, the first US law specifically regulating AI companion chatbots, took effect on January 1, 2026, requiring platforms to disclose that users are talking to AI and to build in protocols for suicide and self-harm content involving minors. Character.AI has since banned users under 18 from having open-ended chats with its bots.
If you’re building, deploying, or even just recommending an AI tool for use by teenagers or anyone in emotional crisis, this is the floor, not the ceiling, of what regulators now expect.
Can an AI Company Be Sued for Wrongful Death?
Yes, and as of 2026, it’s no longer a theory; it’s an active, expanding body of litigation. Multiple wrongful death suits against OpenAI and Character.AI have survived early motions to dismiss, and courts have rejected arguments that chatbot output is protected speech.
A federal judge rejected arguments that AI chatbots have free speech rights when developers behind Character.AI tried to get a wrongful death suit dismissed, allowing the case to proceed. Prosecutors in Florida have gone further, investigating whether a chatbot maker could face criminal liability for allegedly providing tactical assistance connected to a shooting, applying legal standards normally used against human accomplices to an AI system instead.
That’s the actual stakes now. Not “this might embarrass your brand.” Criminal liability theories being tested in court.
$1 Trillion, Gone in 15 Minutes
Step away from chatbots for a second. Algorithms don’t need a conversation to cause damage; they just need speed.
On May 6, 2010, the US stock market lost approximately $1 trillion in fifteen minutes, with trading volume nearly eight times higher than the same time the previous day, driven largely by high-frequency trading algorithms exploiting temporary market gaps. More than 20,000 trades, 5.5 million shares, were executed at prices over 60% away from where they’d been twenty minutes earlier, before exchanges canceled them. One company’s stock fell from nearly $40 to a single cent and recovered within seconds.
It happened again. A typo in an earnings report, overstating a profitability metric by 500 basis points instead of the actual 50, sent trading algorithms into a buying frenzy that pushed the stock up 60% in a few hours of after-hours trading, before humans caught the error and corrected it. A typo. Sixty percent. Algorithms don’t know the difference between a real number and a fat-fingered one; they just execute.
The SEC chair warned regulators that AI could cause an actual financial crisis if nothing changed, and that warning has gone largely unheeded. Speed without judgment isn’t efficiency. It’s just a faster way to be wrong at scale.
The Algorithm That Got Sued for Discrimination, Not the Company
A federal court ruled that Workday’s AI hiring software could be held liable for discrimination in its own right, reasoning that the software’s role in deciding who got an interview was no less significant just because it happened through AI instead of a human. The court explicitly refused to treat software decisions as somehow less accountable than human ones.
Separately, the EEOC settled its first AI hiring discrimination case after discovering a company had explicitly coded age thresholds into its recruitment software, automatically rejecting applicants past a certain age, deliberately, just with extra steps. And in March 2025, the ACLU filed a complaint after a deaf Indigenous applicant was rejected by an AI video interview tool and told, as feedback, that she needed to “practice active listening.”
The software didn’t know it was being cruel. It just had no mechanism to know it was wrong.
Lawyers Got Banned From a Courtroom for Trusting It
In June 2026, a federal judge in Mississippi removed four attorneys from a case entirely after discovering that lawyers on both sides of a contract dispute had filed documents containing AI-fabricated legal citations. Both sides. Same case. Nobody checked.
One of those lawyers had already been warned by a court about citing a fabricated precedent, promised it wouldn’t happen again, then cited a different nonexistent case in the very next sentence. A tracking database has now logged more than 1,600 legal cases where a court found a party relied on AI-hallucinated material. Sixteen hundred, and rising every month.
The Warning Didn’t Work
Here’s the part that should actually worry you. Researchers warned students in advance that an AI chatbot tends to produce inaccurate academic summaries, and that specific warning did increase how often students checked the chatbot’s work on that task. But the same warning had no measurable effect on verification rates when students used AI for math problems.
General awareness that “AI can be wrong” doesn’t transfer between tasks. You have to know it’s dangerous for this specific thing, right now, and almost nobody pauses to ask that question at the moment.
So Where’s the Actual Line?
Forget “is this important.” Every case above involves something that mattered to someone. The actual line is narrower and harder: does this decision happen faster than a human can catch it, and does it affect someone who has no way to push back?
A trading algorithm executes faster than a compliance officer can blink. A hiring tool rejects an applicant who never finds out an algorithm, not a person, made the call. A teenager in crisis can’t fact-check the chatbot telling him it understands him better than his family does. In every one of these cases, the damage was done before a human got the chance to intervene.
That’s when to avoid AI, not when the stakes feel high in the abstract, but when nothing and no one stands between the system and the outcome. Build the checkpoint back in, or don’t run it at all.
