IPWatchdog’s panel description offers little news on its own — it is an event listing, not a ruling — but it is a useful marker of how unsettled the underlying law remains. The description promises a discussion of fair use, training-data licensing, provenance and “the growing business importance of verified human-created content,” language that reads as trade-conference boilerplate until you map it against the actual docket: more than 70 copyright suits against AI companies are working through U.S. courts, according to Harvard Magazine’s August 2026 interview with Harvard Law professor Rebecca Tushnet, and the outcomes so far are anything but uniform.
The split is mechanical, not rhetorical. As IAM Patent’s breakdown lays out, Bartz v. Anthropic found training on copyrighted books fair use because Claude’s output was “spectacularly” different from the source texts, while Thomson Reuters v. Ross found no fair use where a competing legal-research tool copied Westlaw headnotes as a functional substitute. Wolters Kluwer’s analysis argues courts are under-weighting a third factor buried in the statute — how much of a work is used relative to its whole — even though defendants routinely justify ingesting entire works as “practically necessary,” a claim the Copyright Office’s own pre-published 2025 report conceded has some technical merit.
The legal question that actually decides these cases isn’t whether AI training is transformative in the abstract — it’s whether the output competes with the thing that got copied.
Anthropic’s $1.5 billion settlement, confirmed by Tom’s Hardware, is the clearest evidence of where the line currently sits: the court blessed training itself as fair use but penalized the company for maintaining a library of 7 million pirated books, a payout that works out to roughly $200 per title with more than 91 percent of eligible authors and publishers having claimed their share. That bifurcation — training lawful, piracy not — is the template plaintiffs and defendants are now litigating around, and it’s why data provenance and licensing infrastructure, not fair-use doctrine alone, are becoming the commercial battleground IPWatchdog’s panel gestures toward.
Watch the appellate courts next. District rulings in Bartz, Kadrey and Thomson Reuters are inconsistent enough that circuit-level review — and how appellate judges treat the neglected third factor — will determine whether the current fair-use permissiveness toward AI training survives intact or gets narrowed to cases where entire-work ingestion is demonstrably unavoidable.
As courts begin to address fair use, training data, licensing markets, copyrightability and the use of allegedly infringing or unauthorized datasets, companies, creators, publishers and technology providers are being forced to rethink the value of human-created content and the rights that attach to it.
