Start with the premise: you cannot evaluate what you cannot identify. If a text could be written by a person or produced by an algorithm, every judgment about bias, accuracy, or intent collapses until the source is known. We demand that political ads identify funders; we demand that manufactured products list ingredients. The rule is that the public square works only when the origin of information is visible. A label does not suppress AI content, it contextualizes it. The cost of labeling is trivial, while the cost of full anonymity is a citizenry that cannot tell honest testimony from pattern prediction.

The line-drawing objection is a red herring. Nobody would label Grammarly fixing a comma, and no sane law would require it. The statute targets substantive output—paragraphs, images, videos produced by an algorithm without meaningful human authorship. Spellcheck assists; it doesn't author. We already draw messy legal lines: theft requires proving intent, which is inherently fuzzy, yet we enforce it daily. Set the threshold as meaningful human authorship, just as we set an age of majority despite the edge cases. And your enforcement point fails the same way: copyright is violated constantly, but we don't scrap the law. Imperfect enforcement is not an argument against having the rule.

Your Pentagon example actually proves my point: without a legal baseline, no one is accountable for amplification. A labeling law creates a duty on the platforms hosting the bot, just like defamation law targets publishers, not just the original liar. Provenance tools stay voluntary until the law makes them mandatory. You say mandated labels only bind the honest, but that's true of every law; we don't scrap speed limits because some people speed. Evasion is real, but it's an argument for enforcement, not for abandoning the standard.