Microsoft Copilot rarely copies full sentences, firm claims

The official account: Microsoft, defending itself against the New York Times and a cluster of book authors, states in its legal filings that Copilot rarely reproduces full sentences from copyrighted works. The machinery: a word like “rarely” is doing an immense amount of unpaid labour, and the entire dispute turns on what it is being asked to carry.

Consider what “rarely” concedes even as it defends. It does not say “never.” It does not say the system is architecturally incapable of reproduction, only that reproduction is an infrequent output of a process whose normal operation is something else - paraphrase, summary, synthesis. This is a dignified formulation built to survive in a courtroom, where the question is substitution: does the output stand in for the original, such that a reader need not buy the New York Times or the author’s book? Microsoft’s lawyers have chosen their ground carefully. They are not arguing that copying never happens. They are arguing that copying is the exception that proves the system’s normal function is transformative, and that exceptions do not establish a business model.

But watch what this concedes about the underlying convention - because there is always a convention beneath the formal claim, and here it is a statistical one dressed as a legal one. “Rarely” is a frequency claim masquerading as a design claim. It tells you almost nothing about severity. A system that reproduces full sentences from a Times investigation once in ten thousand queries is, on Microsoft’s account, behaving rarely; it may also, on the Times’s account, be functioning as an efficient substitute for a subscription in precisely the cases that matter - the queries put by researchers, students, or rival journalists seeking the exact sentence that carries the exact allegation. Frequency and substitution are different axes entirely, and the genius of “rarely” is that it answers the first while appearing to answer the second.

This is where the authors’ interest and the Times’s interest, though allied in the courtroom, diverge in structure. The Times’s injury, if it exists, is reputational and competitive - a system that can answer “what did the Times report about X” without the click. The book authors’ injury is different: theirs is a claim about the substance of narrative and voice, about whether a chatbot trained on a novel can, on demand, hand a reader something close enough to the original that the purchase becomes unnecessary. Both injuries can be true even if “full sentence reproduction” is statistically rare, because neither injury depends on verbatim copying as its mechanism. Paraphrase that preserves structure, plot, and distinctive phrasing can substitute for a purchase just as effectively as a lifted paragraph - perhaps more effectively, since it evades the very test Microsoft’s filing is built to satisfy.

The confidence dynamics here are worth dwelling on, because they explain why “rarely” is the right word for Microsoft to have chosen rather than a stronger one. A flat denial - Copilot never reproduces protected text - would be falsifiable by a single well-chosen prompt, and litigation is a genre in which single well-chosen prompts are exactly what opposing counsel spend their evenings constructing. “Rarely” is built to survive precisely that kind of adversarial demonstration; it absorbs the occasional damning screenshot without conceding the argument. It is the corporate equivalent of a minister telling Parliament that mistakes were made only in a small number of cases - true, unfalsifiable in the aggregate, and entirely uninformative about whether the small number of cases are the ones anyone cares about.

What the operational analysis reveals, and what the official account obscures, is that this case will not be settled by measuring frequency at all. It will be settled - if it is settled by argument rather than commercial accommodation - by whichever side persuades a court that the right test is frequency of copying or fidelity of substitution. Microsoft has already told you, in a single adverb, which test it intends to win on. The authors and the Times would do well to stop arguing about how rare the parrot’s exact quotations are, and start asking how well the parrot manages, in its own words, never to need them.