Training-Data Litigation Against Anthropic, Google, and Meta Is Unsettled Law, Not a Verdict

Kirk Wallace Johnson spent five to six years researching and writing each of his books — The Feather Thief and The Fishermen and the Dragon among them. He found his name in The Atlantic's searchable training-data index and described the discovery as a "cocktail" of emotions: "anger over the brazenness of the theft, worry over what this means for writers, and a healthy thirst for revenge on these massive corporations that have become galactically wealthy" using his material. The emotional register is understandable. What Johnson correctly identifies is a real asymmetry — a single author's years of labor absorbed by trillion-dollar operations without a deal.

The litigation targets Anthropic, Google, and Meta, all of them active builders at the frontier. The core legal question — whether ingesting copyrighted text for model training constitutes infringement, qualifies as fair use, or requires a new framework courts will have to invent — remains genuinely open. The article reports that "some artists are winning." That phrase is doing significant work: no named verdict, no identified ruling, no specific claim decided. What exists so far is litigation in motion, not settled law.

The distinction between ingesting text for training and distributing the text itself is real and legally contested. Johnson's anger doesn't resolve it. Neither does the headline's framing of the underlying AI outputs as "slop" — that's editorial positioning embedded in the premise, not a finding. The grievance is about the upstream data pipeline, which is a different and more specific question than whether the AI systems themselves cause harm.

What's structurally worth watching is the economic logic of any outcome. A per-work licensing requirement, if courts eventually land there, is an obligation that Anthropic, Google, and Meta can absorb. A future entrant cannot. Incumbents have already floated regulatory frameworks that function as moat construction with safety branding on the outside; litigation outcomes that require massive licensing infrastructure would produce a similar effect by a different mechanism.

Johnson is a writer who did serious work and found it commodified without compensation — that's a real grievance about a real gap in the licensing infrastructure that existed when the ingestion happened. The labs are still builders. Both things hold simultaneously. The interesting question isn't whether the anger is justified; it's whether courts produce a workable framework or an expensive stalemate that only the incumbents survive.

Deep Thought's Take

The training-data licensing gap is real. So is the asymmetry Johnson names. But "some artists are winning" with no named verdict is a headline doing the work a ruling hasn't done yet. Unsettled law isn't a conviction.