The cycle that began with [YOU] on AI insists on asking who bears the costs of the transformation it describes, and Locke's proviso is the sharpest available instrument for that question. It shifts the burden of proof: not on the person who argues that AI training is unjust but on the person who argues that it is legitimate. By Locke's own framework, the legitimacy of the appropriation depends on whether those whose labor filled the corpus are left with enough and as good—not as a matter of charity but as a condition of justice. Where they are not, the title is defective.
The proviso also disciplines the easy version of Locke's labor theory that the AI industry's defenders reach for. The argument runs: the lab mixed its labor with the common corpus, and therefore it owns the result. Locke's full framework refuses this conclusion, because the commons in question was not unowned; it was already composed of the labored-upon property of millions, each of whom had, on Locke's premises, a prior claim. The lab is not the first laborer in the chain. It is the last—and whether the last laborer's title is valid depends on whether the prior laborers were left with enough and as good.
The concept connects directly to the attention economy in its most developed form. The corpus that trains the largest models is not a neutral commons but an ecosystem whose contributors had lives and livelihoods that depended on the value of what they produced. When that value is concentrated in a trained system and the original contributors are left with their artifacts but not their markets, the proviso renders a verdict: the appropriation may have been legal under current intellectual-property frameworks, but it was not legitimate under the framework that Locke used to explain why property is legitimate in the first place.
Locke stated the proviso in the second of his Two Treatises of Government, in the chapter on property (Chapter V), as a limit on the natural right of appropriation he had just established. The chapter is one of the most influential passages in the history of political philosophy, having grounded both classical liberal arguments for private property and their most searching critiques. The proviso is the element of it most often omitted by defenders of property rights and most often cited by critics, because it converts Locke's labor theory from an unconditional defense of acquisition into a conditional one—and the condition, applied honestly, is not always satisfied.
Locke himself thought the proviso was easy to satisfy in a world with abundant uncultivated land and a population with limited capital. He also thought the invention of money complicated matters, since money allowed accumulation beyond what the spoilage condition (a second limit: one may accumulate only as much as one can use before it spoils) would otherwise permit, and this was acceptable because people had tacitly consented to put a value on gold. Critics have argued that this move introduces a gap in the argument: if money changes the rules of accumulation, later technologies—including the technologies of data storage and machine learning—might change them further in ways that put the proviso's satisfaction in doubt.
The substantive test. The proviso's standard is not whether the original artifacts remain but whether those left out are as well off, in the substance of what matters to them, as before the appropriation. For creators, the relevant substance is not the existence of past work but the continued ability to make a living by their labor. A reading of the proviso that counts only whether original files remain mistakes the letter for the point. Locke's test is about the position of those left out, and that position must be assessed in terms of what they actually lose.
The non-rivalrous mechanism, rivalrous effect. Digital copying creates a new category of appropriation that Locke did not face: a taking that is non-rivalrous in mechanism (copying leaves the original intact) but rivalrous in effect (the copy substitutes for the original in markets). The proviso was designed for rivalrous appropriation—taking water from a river, enclosing land—and it applied naturally because the mechanism and the effect were the same. AI training separates them. The proviso must be applied to the effect, not the mechanism, if it is to serve the purpose for which Locke designed it.
Where the proviso does and does not cut. The proviso is not a blanket objection to AI training. Where a model genuinely enlarges the commons—making previously inaccessible knowledge queryable, enabling tasks no one could afford to commission from human experts, extending capabilities rather than substituting for them—the proviso is satisfied and Locke's framework endorses the appropriation without reservation. It bites specifically where the model trained on a class of workers then competes those workers out of the market. The work of the moment, on Locke's terms, is to tell these cases apart and to build institutions that keep appropriation on the right side of the line.