"They have sown the wind, and they shall reap the whirlwind." — Hosea 8:7
Somewhere in the training data of a model you can rent by the month, there is a piece of your work. A blog post. A product description. A thread you wrote at midnight because you had something to say. You were not asked. You were not paid. You were not told. And now the system built on top of it can produce a thousand imitations of your voice for a fraction of a cent each.
This is the part of the AI story that the people selling AI would prefer you experience as weather — something that simply happened, an inevitability, the price of progress. It was not weather. It was a series of choices, made by specific companies, that are now being adjudicated in specific courtrooms. And the verdicts are starting to come in.
What "We Respect Creators" Costs
Every major AI company has a public position on creators, and that position is warm. They value human creativity. They want to partner with the creative community. They are building tools to empower writers and artists and developers, not replace them. You have read the blog posts. They are very well written.
Now hold that warmth against the record. In 2024 a group of authors sued Anthropic, alleging the company trained its models on millions of copyrighted books. In June 2025 a federal court reached a split decision that is worth understanding precisely: training a model on books could qualify as fair use — but the way the company had obtained many of those books could not. The judge's order found that Anthropic had pirated more than seven million copies. By September 2025 the company had agreed to settle for roughly $1.5 billion — one of the largest copyright settlements in American history — working out to around $3,000 for each of an estimated 500,000 books covered. The New York Times's suit against OpenAI and Microsoft, built on a similar theory about journalistic work, is widely expected to settle in 2026.
Notice what the settlement is and is not. It is an admission that the input had value — enough value that paying for it after the fact runs to ten figures. It is not a system that asked you first. The model got built either way. The compensation arrived only because someone with the resources to sue forced it into existence. For every author inside that class, there are millions of writers, marketers, photographers, and developers whose work is in the same training sets with no settlement, no class, and no check.
The Quiet Inversion
Here is the move, stated plainly, because the industry will never state it plainly for you.
Your work was treated as a free raw material — "data," the new oil, there for the taking. The model trained on it was then treated as valuable private property — licensed, metered, protected by terms of service and trade-secret law. The same content is public commons on the way in and proprietary asset on the way out. The value did not disappear in that transformation. It moved. It moved from the people who made the work to the people who collected it.
And then the loop closes. The tool trained on your output is sold back to you as a way to produce more output faster — which is to say, as a reason your particular skill is now worth less in the market. You supplied the training data for your own depreciation. You were the input, the customer, and the casualty, in that order.
Why the Word "Inevitable" Is Doing So Much Work
The most useful word in this entire debate, for the companies, is inevitable. If the absorption of all human creative output into a handful of private models is simply going to happen no matter what, then resistance is naïve, licensing is friction, and your only rational option is to adopt the tools and stop complaining.
But inevitability is a claim, not a fact, and it is a claim that conveniently benefits exactly one side. The settlements prove the point: when consent and compensation were enforced — by a court, with teeth — the industry did not collapse. It paid. The business model that supposedly could not survive paying creators turned out to be perfectly capable of writing a $1.5 billion check. What was presented as technologically impossible was merely expensive, and "expensive" is a negotiation, not a law of physics.
The same is true for you, at your scale. Your individual leverage is small. Your collective leverage is not, and the people who say otherwise are the ones who benefit most from you believing it.
What This Means for an Operator
If you make things for a living — words, images, code, strategy, anything — you are both a supplier and a competitor to these systems, whether you signed up for that or not. A few moves follow from taking that seriously.
Know what you're feeding. Read the terms of the platforms you publish on and the tools you draft in. Some grant themselves broad rights to use your content for training; some don't. The difference is in the boring paragraphs nobody reads, which is precisely why they're written the way they are. You cannot control everything, but you can stop volunteering.
Keep what only you can make. The work that is hardest to absorb into a model is the work that draws on things a model does not have: your direct experience, your proprietary data, your relationships, your judgment in a specific room with specific people. Generic output is the most replaceable thing you produce. Specificity is the moat.
Treat your archive as an asset, not exhaust. The body of work you've built has value — demonstrated, now, in court, at roughly $3,000 a book. Catalog it. Own the canonical copies. Be deliberate about where it lives and who gets rights to it. The companies certainly know what it's worth.
Don't accept "inevitable" as a term of service. When a vendor frames the surrender of your rights as the unavoidable cost of using modern tools, that is a sales position, not a fact. Tools exist that take less. The market for them grows precisely as people start asking the question.
The Whirlwind
None of this is an argument against the technology. The models are genuinely useful, and pretending otherwise helps no one. It is an argument against the terms — against a deal in which the people who supplied the value are the last to be considered and the first to be displaced, all narrated as progress that simply happened.
It did not simply happen. It was built, on a specific raw material, much of which was taken. The courts are now putting a price on that material, and the price is not small. The lesson for everyone who makes things is the same lesson the settlements teach: your work has value the moment it leaves your hands. The only question is whether you act like it before someone else does.
Sources: NPR and Authors Guild reporting on the Bartz v. Anthropic settlement ($1.5 billion; ~$3,000 per book; ~500,000 books); the June 23, 2025 court ruling on training fair use versus pirated acquisition (7M+ copies); reporting on The New York Times v. OpenAI and Microsoft; Norton Rose Fulbright and AI Business overviews of 2026 AI copyright litigation.
