Andersen v. Stability AI Trial Puts the “Model-as-Copy” Copyright Theory Before a US Jury for the First Time

What’s Actually Being Tested

Most AI copyright litigation to date has centered on outputs — whether a specific image, text, or song generated by an AI system reproduces a copyrighted work closely enough to infringe. Andersen v. Stability AI tests a different and more foundational theory: that the trained model itself, as a data artifact containing statistical traces of the copyrighted images used to train it, constitutes an infringing copy — independent of whether any particular output resembles any particular input work.

That distinction matters enormously for how the case could reshape the industry. An output-infringement finding affects specific generations that too closely resemble specific copyrighted works — a problem addressable through filtering and output controls. A model-as-copy finding would implicate the training process itself, meaning any model trained on copyrighted material without a license could be found infringing regardless of what it ultimately generates. The case, filed under docket number 3:23-cv-00201, has already seen its DMCA Section 1202 claims dismissed earlier in the proceedings, narrowing the surviving claims but not eliminating the core theory now headed to a jury.

Why a Jury Verdict Changes the Calculus

Copyright disputes involving AI training have so far been resolved primarily through settlements, licensing deals, or judicial rulings on narrow procedural questions — not through a jury actually weighing the merits of the model-as-copy theory. A jury verdict, unlike a settlement or a motion ruling, sets a factual and legal marker that other courts, plaintiffs, and defendants can point to directly. Win or lose, the outcome becomes a reference point every subsequent AI copyright case has to account for.

For AI companies, a verdict against Stability AI on the model-as-copy theory would be significantly more disruptive than an adverse output-infringement ruling, because it would call into question the legality of the training process for any model built on unlicensed copyrighted data — not just the specific model at issue in this case. For rights holders and artists, a favorable verdict would validate a legal theory that has, until now, remained largely untested at trial.

The Broader Litigation Landscape

Andersen v. Stability AI is not proceeding in isolation. Related cases working through similar questions include Disney, Universal, and Warner’s suit against Midjourney over image and video generation, currently in discovery, and Sony’s litigation against Suno and Udio over music generation, with a ruling on the core fair-use question anticipated in summer 2026. Getty Images has pursued parallel claims against Stability AI in both the UK, where the High Court rejected Getty’s secondary copyright claim in November 2025 and the case is now under appeal, and in the US, where trademark and unfair-competition claims advanced in April 2026.

This cluster of cases collectively covers most of the major generative AI modalities — images, video, and music — and most of the major legal theories, from direct infringement of outputs to the model-as-copy theory now before an Andersen jury to secondary and trademark claims. Whatever else happens, the Andersen verdict will not settle every open legal question around AI training data, but it will be the first data point from an actual jury on the theory most AI companies have treated as their deepest legal exposure.

What to Watch For

The jury’s verdict, whenever it lands, will be closely parsed for whether it addresses the model-as-copy theory specifically or resolves the case on narrower grounds. A verdict that avoids the core theory — for instance, one that turns primarily on the surviving claims rather than a clean ruling on whether the trained model itself infringes — would leave the central legal question about AI training largely unresolved even after the trial concludes, pushing a definitive answer to one of the parallel cases still working through the courts.

Frequently Asked Questions

Sources & Further Reading

• AI Copyright Lawsuit Tracker 2026 — Rinzara