Competition Isn't Theft

Call competition theft often enough and eventually you stop having to make distinctions.

That is what makes J.A. Konrath’s recent AI rhetoric so revealing. The legal confusion is sloppy, sure. But the more interesting part is the story underneath it: the writer as victim, the machine as killer, and the market itself recast as a crime scene. It is vivid language. It is emotionally legible. It also collapses several very different claims into one inflammatory word.

Piracy is one thing. Copyright infringement is another. Passing off work under someone else’s name is another. Style imitation is another. Market competition is another still. If you jam all of them into “theft,” you get a better slogan and a much worse argument.

That distinction is not just semantic hair-splitting. On June 23, 2025, Judge William Alsup drew exactly this kind of line in the Anthropic case. He held that Anthropic’s use of books to train Claude and its precursors was fair use on the facts before him, while also holding that pirated library copies were a separate problem.

“Such piracy of otherwise available copies is inherently, irredeemably infringing…”

Judge William Alsup, Bartz v. Anthropic PBC, June 23, 2025

That matters because it breaks the lazy shortcut so much of this discourse depends on. “Some AI companies used pirated copies” is a real accusation. “AI training is theft” is a much broader claim, and the law is not treating those two statements as interchangeable. The judge did not bless piracy. He distinguished between allegedly transformative training use and the separate act of building a library out of unlawfully acquired copies.

A different line showed up on February 11, 2025 in Thomson Reuters v. Ross Intelligence. There, the court rejected Ross’s fair-use defense in a dispute over Westlaw headnotes used to help build a competing legal-research product. The facts were different, the market relationship was different, and the result was different. Which is exactly the point. Recent decisions are not saying “everything AI does is fine” or “everything AI does is theft.” They are trying, imperfectly but visibly, to separate one kind of use from another.

The style complaint gets even messier.

Copyright protects expression. It does not hand out deeds to cadence, sentence pressure, or the general feeling of how a writer sounds on a good day. The Copyright Office’s digital-replicas report makes the boundary fairly plain. It acknowledges that where a style is closely tied to identity, some existing right-of-publicity, unfair-competition, or deception claims may come into play. But it also says the Office does not recommend making style itself protected subject matter under a federal digital replica law at this time.

That is a much cleaner frame than the panic usually allows. If someone uses your name to sell garbage you did not write, that is a real harm. If a system generates output that copies protectable elements of a specific work, that may be infringement. If a model imitates broad stylistic patterns without crossing those lines, the argument gets much weaker much faster. Style can be commercially important. It can be personal. It can even be bound up with identity. But it is still not the same thing as owned property in the copyright sense.

And once you see that, the emotional center of the complaint starts moving. The real sentence in all of this is not “they stole from me.” The real sentence is “I can’t compete.”

That fear is understandable. If a tool lowers the price of competent prose, familiar structure, and serviceable stylistic imitation, then some kinds of writing lose scarcity. That is a market shock. It may be brutal. It may be unfair in practice. It may justify new licensing regimes, stronger disclosure rules, or better protections against deception and memorization. But it is still a different category of problem than theft.

Konrath himself used to understand that distinction more clearly than he seems to now. Back in September 2023, writing about AI and the Authors Guild lawsuit, he asked a sharper question than the one he appears to be asking today:

“Does the Writers Guild really fear theft? Or competition?”

J.A. Konrath, “English Language Sues Bestselling Authors for Using English Without Permission,” September 29, 2023

That line lands because it names the actual pressure. It also boomerangs. Once you admit the deeper fear is competition, the melodrama starts losing altitude. You no longer need to pretend every market threat is a rights violation. You can talk honestly about substitution, discoverability, pricing pressure, licensing, and the collapse of old advantages.

That is why the “one million more books” talking point does not prove what people want it to prove. Publishers Weekly reported on March 17, 2026 that U.S. ISBN output topped four million in 2025, up 32.5 percent from 2024, with self-published titles driving most of the increase. But the explanation offered there is broader than “AI wrote them.” Bowker’s cited factors include the growing availability of tools, easier publishing, distribution and marketing access, and the fact that multiple formats each require their own ISBN. The number is real. The causal leap from that number to “AI-assisted writing caused the flood” is still an inference.

More to the point, even if AI did help increase output, abundance is not evidence of theft. A glut of mediocre books may create a discoverability problem. It may cheapen certain market signals. It may make curation more valuable. None of that turns competition into larceny.

So it helps to call each thing by its proper name.

If a company builds a dataset from pirate libraries, call that piracy.

If a model reproduces protected passages or copies protectable expression too closely, call that infringement.

If someone sells AI-generated work under a real author’s name, call that passing off, deception, or a digital-replica and identity problem.

If cheap imitations flood the market and pressure working writers, call that competition.

Writers are allowed to hate competition. They are allowed to resent the tools that intensify it. They are allowed to argue for stronger rules, better contracts, clearer consent, and compensation structures that reflect the actual economics of training and substitution. What weakens the case is the insistence that every one of those concerns must be smuggled back under the word “theft” in order to feel morally serious.

Cultural production has always moved through study, imitation, variation, and recombination. Writers learn by absorbing other writers. Storyforms recur because narrative pressure recurs. Voice evolves through influence long before it becomes recognizable as voice. AI does not abolish that older reality. It industrializes parts of it. That scale change may justify policy intervention. It does not automatically convert competition into murder.

And that is why this whole argument keeps sounding overheated. Beneath the accusation is a much older panic, one the publishing world has seen before whenever tools, formats, or distribution change the terms of entry.

What if I am not losing my work?

What if I am losing my advantage?

Sources

1. Bartz v. Anthropic PBC, Order on Fair Use, June 23, 2025
2. Thomson Reuters Enterprise Centre GmbH et al. v. ROSS Intelligence Inc., Memorandum Opinion, February 11, 2025
3. U.S. Copyright Office, Copyright and Artificial Intelligence, Part 1: Digital Replicas
4. Publishers Weekly, “Book Output Topped Four Million in 2025”
5. J.A. Konrath, “English Language Sues Bestselling Authors for Using English Without Permission”