First truth: a company should not be forced to provide technology for a use it deems unacceptable. Second truth: a private company should not be the sole judge of what a democratically elected state may do. Both seem solid, and they contradict each other. That is precisely the crux of the matter.
We covered the dispute between a lab and the Pentagon, then noted this morning an absence from the list of selected vendors. Let us take the time to address the underlying question.
The case for refusal
It is an old and respectable one. Whoever makes something bears a share of responsibility for what it enables. This is the principle behind conscientious objection clauses, an architect's refusal to design certain buildings, or a researcher's refusal to work on certain programmes.
Three arguments support it in our case.
Causal proximity. An AI vendor is not selling a neutral tool: it is selling a system whose capabilities it has defined, and whose potential uses it knows precisely.
The absence of a framework. We regularly document that rules arrive after capabilities. In the meantime, if the manufacturer does not set a limit, no one else does.
Irreversibility. Some uses, once installed, cannot be withdrawn. This is what we noted in our article on future generations: prefer what can be undone.
The counter-argument, and it is serious
It cannot be reduced to a defence of commerce, and that is what makes it uncomfortable.
Legitimacy. A company is elected by no one. When it decides a use is unacceptable, it substitutes its judgement for that of institutions which do proceed from a democratic process. One may find that judgement better; it is not more legitimate.
Selectivity. These refusals are rarely systematic. A company that refuses one use and accepts another, comparable one, is exercising a filter whose criteria are neither public nor contestable.
Displacement. A refusal does not eliminate the use: it changes the supplier. We saw this morning — the deployment went ahead with others. The refusal therefore cost a contract without preventing anything.
This question should not be framed in these terms. If we are debating a company's right to refuse a client, it is because no public rule exists stating what is permitted. In a regulated domain — conventional weapons, civil nuclear power, health products — a manufacturer does not set its own ethical policy: it applies a framework that has been debated, written down and overseen. The debate over private refusal is a symptom of a void, not a solution to it.
What refusal still achieves
Despite its limits, a public refusal produces two real effects.
It makes visible a question that would otherwise have stayed buried in confidential contracts. Without this dispute, no one would be publicly debating what a vendor may demand of a public client.
And it sets a precedent. Legal frameworks are often built from conflicts that force a judge to settle a question the legislator had not anticipated.
From this standpoint, a refusal that fails commercially can succeed politically. That is not a consolation; it is a function.
What to take away
Our position, and it will not fully satisfy anyone: private refusal is a legitimate and insufficient stopgap.
Legitimate, because in the absence of rules someone must set limits, and it is better for one actor to publicly take a stance than for no one to do so. Insufficient, because it rests on the judgement of companies that nothing obliges to maintain it, especially when it comes at a cost.
What seems most solid to us is that these refusals should be understood as alerts rather than solutions. When a company refuses a use, it signals that a question has not been settled collectively. The right response is not to praise or condemn it, but to settle the question.