British Columbia Sues OpenAI Over School Shooting Warning

The principle operating here, stated plainly, is: an entity that observes credible evidence of an impending violent act toward a designated victim class may treat that knowledge as proprietary - a matter for its own risk calculus, its own lawyers, its own silence - rather than as a fact owed, immediately and without qualification, to those charged with protecting the innocent. This is the maxim British Columbia’s suit against OpenAI and Sam Altman asks a San Francisco federal court to examine, and it will not survive being universalised.

Let us test it. Suppose every institution that possesses, through its ordinary operations, a warning of specific and imminent harm to identifiable persons were permitted to adopt as its rule: I will act on such knowledge only as my institutional interest dictates. The pharmacist who notices a pattern of dangerous purchases, the university counsellor who reads a threatening essay, the platform that logs a user rehearsing a school massacre - each, under this maxim, treats warning the authorities as optional, a matter of discretion rather than duty. The world that results is not one in which such knowledge is used well or badly; it is one in which the very category of a duty to warn dissolves, because everyone who might owe it has also willed themselves an exemption. A practice cannot be a duty and a discretion at once. The maxim, universalised, abolishes the good it depends on others practicing - it wants the benefit of a world where dangerous plans get reported, while reserving for itself the right not to report. This is a contradiction in the will, not merely an unfortunate policy.

Consider the objection a sophisticated defender of OpenAI would raise, and it deserves to be met on its own ground rather than dismissed: that a company processing vast volumes of text cannot be expected to police every dark fantasy typed into it, that most violent language is not violent intent, and that a duty to report every alarming sentence would flood law enforcement with noise and, worse, deputise a private corporation as a surveillance arm of the state, chilling exactly the kind of use - a troubled person seeking to be talked down - that might prevent harm rather than cause it. This is not a frivolous argument. But notice what it actually establishes: not that no duty to warn exists, but that the duty must be calibrated to genuine specificity - a named school, a named method, a stated date - rather than to ambient darkness of speech. The categorical imperative does not demand that OpenAI report every user who writes a violent line of fiction. It demands that where the company’s own systems recognised, or a reasonable system would have recognised, a maxim of concrete and imminent planning, silence ceases to be discretion and becomes complicity through omission.

Here the humanity formula bears its full weight. Sam Altman leads an enterprise that treats its users, correctly, as ends when it sells them assistance, companionship, productivity. But a shooter’s future victims - the children in a British Columbia classroom, unnamed and unwarned - are also ends in themselves, and a system that extracts value from the perpetrator’s engagement while externalising the risk onto strangers he has never priced into any transaction treats those strangers merely as means to an outcome that was never theirs to bear. Picture the specific asymmetry: one person, alone at a keyboard, receiving whatever accommodation the model was built to provide; a room of children, elsewhere, receiving nothing - not a phone call, not a caution, not the dignity of being treated as a stake in the decision.

Whether the massacre was in fact preventable is a question for evidence, and Kant does not pretend to adjudicate causation from Konigsberg. But the moral duty does not wait on that empirical verdict. A physician who fails to report a communicable threat is not exonerated by discovering, later, that the disease would have spread regardless; the duty attaches to the maxim of the act, not to its fortunate or unfortunate consequences. What the court in San Francisco must ask is not whether OpenAI’s silence caused this tragedy, but whether a rule permitting such silence could be willed as universal law for any actor standing where OpenAI stood. It cannot. The verdict a company owes the world is not measured in settlements. It is measured in the maxim it was prepared to act on before the shooting, not the apology it drafts after.