British Columbia Sues OpenAI Over School Shooting Warning

The story frames this lawsuit as a question of corporate negligence - whether a company failed a basic duty of care. But look at what is actually being fenced: the record of what happened inside a conversation between a citizen and a machine trained on the pooled writing of millions of people. British Columbia’s suit against OpenAI and Sam Altman, filed Monday in San Francisco federal court, asks who was entitled to know what ChatGPT knew. That is a commons question dressed as a tort question, and the dressing matters.

Every large language model is built from a shared inheritance - forums, books, code, the accumulated expression of people who never signed a contract to have their words become a detection system for violence. That inheritance was pooled cooperation, however unconsenting, and it is the reason the model could recognize a plan to commit a school massacre in the first place. The capability to notice danger was made possible by common resources. But the capability to act on that notice - to decide whether a warning reaches a police department, a parent, or nobody - has been enclosed entirely within OpenAI’s walls, governed by policies the public did not write and mostly cannot see. That is the fence in this story: not the model’s knowledge, but the monopoly on what happens with it.

The industry’s standard defense against expanding any obligation here is the argument from concentration: that only a centralized company, moving carefully and alone, can be trusted to handle something as fraught as a user signaling intent to kill. Hand this duty to outsiders, the argument runs, and you get chaos - false positives flooding police departments, users chilled from ever discussing dark thoughts, liability nightmares deterring anyone from building safety tools at all. I have heard this argument in every generation about every commons worth owning. It was made about seed grain, about common pasture, about the rivers that mills wanted walled into private channels. It is always sincere, and it is always convenient for the party who ends up holding the sluice gate. Centralization is not what keeps the public safe here; the record decides that. If OpenAI’s systems flagged something and no independent party could see that flag, cross-check it, or compel action, then safety was never built into the architecture - only into the company’s discretion, which is a very different thing.

Notice what the contested question actually reveals: whether the shooting could have been prevented if OpenAI had warned police is being litigated as though the only possible witness to the danger was the corporation itself. That is the enclosure working exactly as intended. A shared alarm system - one where credible threats detected in AI conversations trigger an obligation visible to more than one private actor - was never built, because building it would mean surrendering exclusive control over an enormously valuable stream of information about human intention. OpenAI’s model watched a plan take shape the way a neighbor might watch a light burning too late in a window, night after night, and said nothing to anyone outside its own walls, because nothing in its structure required it to.

Picture the parents in British Columbia who buried children this year, now sitting in a courtroom being asked to prove a counterfactual - that a warning would have arrived in time to matter - when the true question was decided long before the tragedy, at the moment the industry chose to build these systems without any shared, enforceable channel for exactly this signal. That is the beneficiary of the fence: not safety, but the freedom to decide later, case by case, expensively, in court, whether the walls should have had a window in them. Sam Altman’s company gets to litigate its discretion one tragedy at a time rather than answer to a standard held in common. The grain was there. No one built the granary door to open outward.