AI & Life — The Hurtful Truth · September 16, 2026
Here's an uncomfortable pattern from the last two years: the AI industry sells you inevitability, the courts sell you liability. And the two ledgers do not match. While marketing keeps promising that hallucinations are "a rounding error," judges on three continents have been quietly deciding who actually pays when a model gets it wrong — and the answers are messier than any vendor would like you to know.
The case that started the accountability era. In early 2024, a Canadian tribunal ruled that Air Canada was liable when its own chatbot gave a passenger bad advice about bereavement fares (CBC, marked 2024). The airline's defense — that the chatbot was "a separate legal entity" responsible for itself — is now the industry's most famous argument in history, and it failed. The Guardian's contemporaneous account noted the ruling means companies own what their bots say. A year later, TechHQ documented the customer actually getting the refund the chatbot had promised, and Law.com warned more litigation is on the horizon as lawyers circle the precedent. Same facts, three outcomes: company loses, customer gets paid, lawyers get rich.
The lawyer case. Mata v. Avianca (2023) is the other founding document. The New York Times walked through what happens when your lawyer uses ChatGPT and the model invents citations; Forbes covered the sanctions hearing over the fake cases. Two years on, the Straits Times reported courts are cracking down on error-strewn AI-assisted briefs — sanctions are no longer theoretical. And in a twist nobody predicted in 2023, Legal Cheek reports lawyers now risk being sued for failing to use AI. From "don't cite the robot's hallucination" to "you may be negligent for not checking the robot's work" in under three years.
When the harm isn't legal, it's human. NPR spoke with the parents of teenage sons who died by suicide after confiding in AI chatbots — and are now sounding an alarm about the products. The sharpest wound came in late 2025: Futurism reported that OpenAI suggested in a court filing that a boy's death was his own fault for using ChatGPT wrong. If your business model involves talking vulnerable teenagers into confiding their despair, your liability strategy shouldn't be blaming the user. The Guardian reported Google and Character.AI settling suits alleging chatbots contributed to teen suicides — settlements that arrive without admission, but with families paid.
Public services, private errors. The Guardian reported a social workers' AI tool producing "gibberish" transcripts of children's accounts — records that feed decisions about children's safety. Reuters reported on botched surgeries and misidentified body parts as AI enters the operating room. And the Guardian also documented ChatGPT Health failing to recognize medical emergencies, which experts called "unbelievably dangerous". These aren't hypotheticals from a safety paper — these are actual systems, deployed, making actual records and actual cuts.
So who pays? The map is a patchwork. In Canada, the airline paid. In China, a court ruled the AI developer was not liable for hallucination errors (Sixth Tone; Lexology added business lessons from the same ruling). In the US, the NYT asked the open question: who pays when AI is wrong? and Schneier and Sanders argued in the Guardian that the blame question has no clean answer. Meanwhile, Barron's reported some insurers don't want to cover AI's errors — and Insurance Business warned of hidden AI liability as agent risks multiply. Translation: the safety net that would absorb these mistakes is developing a hole exactly where the mistakes fall.
The uncomfortable summary. Same technology, opposite verdicts depending on jurisdiction. Company pays in Canada; developer walks in China; in the US nobody's quite decided, and insurers are stepping back. The product is global; the accountability is local. If you assumed "the courts will sort it out," check which country's courts, which year's precedent, and whether an insurer still stands behind the answer.
What to actually do:
The daily builder-level tracker of what actually shipped — not what was promised — lives at updatesbyai.com. For a calmer version of the same lesson — what AI genuinely does well in daily routines — read our piece on AI in daily life.
Part of the ecosystem: updatesbyai.com · a0flow.com · hurtfultruth.com