Unnecessary — Issue 030
The Technology Moved Fast. The Legal
System Is Still Catching Up, Slowly, on Purpose.
The Getty v. Stability AI ruling teaches a broader lesson about what happens when deployment outpaces the rules governing it.
There's a specific, uncomfortable gap that shows up whenever a powerful new technology deploys faster than the legal and regulatory systems built to govern it — a gap where real harm and real risk accumulate on both sides while the slower institutions work through the actual questions at their own deliberate pace.
Generative AI models have been trained on copyrighted creative work and commercially deployed for years already. The actual legal question of whether that training was lawful is, as of the biggest court ruling on the subject so far, still formally unresolved — the UK's landmark Getty Images case sidestepped the core question entirely, leaving it for a future court to decide.
That gap between technology and law is worth understanding clearly, because it's not unique to AI art — it's a recurring pattern worth recognizing wherever it shows up next. Here's the full picture of where this specific fight actually stands.
What the Court Actually Ruled, and What It Deliberately Didn't
The single most important detail in this entire case is one that's easy to miss inside the "Getty loses" headlines: the central, most consequential legal question — whether training an AI model on copyrighted images constitutes copyright infringement in the first place — was never actually decided by this court at all. Getty had originally brought that claim, the "primary infringement" argument, but withdrew it during the trial itself, after considerable evidence had already been submitted and cross-examination had already occurred. The stated reason was jurisdictional: the actual training of Stable Diffusion had taken place entirely outside the UK, largely putting it beyond the reach of UK copyright law regardless of how the underlying legal question might eventually be resolved.
Crucially, the judge specifically ensured the door remained open on that core question for future cases, requiring language in the final order making clear that while Getty's primary infringement claim was abandoned, it was never actually adjudicated on its merits by the court. That's a genuinely significant distinction, and one that both sides of the current AI-copyright debate have real incentive to obscure in their public messaging: this wasn't a ruling that training AI on copyrighted work is legal. It was a ruling that this particular UK court, on this particular set of facts, couldn't or wouldn't decide that question — leaving it explicitly available for a future court, quite possibly on appeal from this exact case, to resolve properly.
Getty was granted permission to appeal in December 2025, and legal analysts covering the case describe the upcoming Court of Appeal hearing as a genuinely significant opportunity to provide the clarity this initial judgment deliberately avoided. The war, in other words, isn't over. It's paused, on a technicality, exactly where the hardest question still sits unanswered.
The Real Verdict Is Coming From California, Not London
While the UK case has generated the most legal commentary so far, the case that will likely produce the first genuinely dispositive answer on the core "is AI training copyright infringement" question is happening in an American courtroom, and it's scheduled for a jury trial — not just a judge's written ruling — in September 2026.
Andersen v. Stability AI, a case brought by a group of visual artists rather than a single corporate plaintiff, is heading toward what current legal tracking describes as the first U.S. verdict specifically on what's being called the "model-as-copy" theory — essentially the identical core question the UK court sidestepped, but this time facing a jury rather than a single judge, under U.S. copyright law rather than UK law. That distinction matters enormously: U.S. copyright law's fair use doctrine operates under a genuinely different framework than UK copyright law, and a jury verdict carries different precedential weight and public visibility than a single judge's written opinion. Legal trackers covering the AI-copyright landscape describe this trial as running in parallel with a broader cluster of related American cases — including litigation brought against Stability AI, Disney, Universal, and Warner Bros. over similar training-data claims — all converging toward late 2026 and 2027, suggesting the genuinely defining legal precedent on this entire question is still roughly a year or more away from being set, regardless of how the UK appeal resolves in the meantime.
The Governments Are Already Answering the Question the Courts Won't
While litigation grinds through appeals and pretrial motions, government policy processes have moved considerably faster, and the results reveal a genuine, hardening international split on the underlying question — which means the eventual legal landscape for AI-generated content may end up varying dramatically by jurisdiction, regardless of how any single court case resolves.
The UK government ran a public consultation specifically on potential copyright law reforms addressing AI training, and the results were about as one-sided as a contested policy question ever produces: over 11,500 responses were submitted, with an overwhelming 88% majority favoring a requirement that AI companies obtain a license before using copyrighted works for training. That's a genuinely striking public policy signal, and it puts real pressure on UK lawmakers to act through legislation rather than waiting indefinitely for the courts to fully resolve the question through case law — precisely the kind of dynamic where a government moves to settle by statute what litigation has proven too slow or too narrow to resolve cleanly.
Germany's Munich Regional Court has already moved further and faster than the UK, ruling in favor of GEMA — the German musical royalty collecting society — in a case establishing meaningful copyright protections against AI training on musical works, a genuinely more creator-favorable outcome than the UK's Getty ruling produced. China, by contrast, has moved in the opposite direction on a related but distinct question: Beijing's Internet Court ruled in September 2025 that AI-generated images can themselves be copyrighted under Chinese law, provided sufficient "creative effort" went into producing them — directly contradicting the U.S. Copyright Office's own position, established in a formal January 2025 report, that purely AI-generated outputs are not copyrightable at all under U.S. law, though human modifications, arrangement, or curation of AI outputs can still qualify for protection.
That divergence is worth sitting with directly: the three major jurisdictions currently shaping global AI policy — the UK, the US, and China — are landing in three meaningfully different places on questions that are directly relevant to anyone using AI-generated content commercially, anywhere in the world, given how easily digital content crosses borders.
What This Actually Means If You Use AI-Generated Content
For anyone using AI image or text generation tools commercially — which, given how deeply these tools have already been integrated into everyday content and design workflows, is an enormous and rapidly growing population — the current legal landscape produces a genuinely uncomfortable, practically important set of open risks rather than clear answers in either direction.
Current U.S. legal guidance is explicit on one point: if the AI vendor whose tool you used is later found, in ongoing or future litigation, to have infringed on training data, your own use of that tool's output could constitute what's described as a "downstream act" of infringement — even though you, the end user, never personally scraped or copied anything yourself. There is a specific statutory protection that matters here: an "innocent infringement" damages floor of $200 per work under U.S. copyright law, which meaningfully caps the financial exposure for someone who used an AI tool in good faith without any knowledge that its training data might later be ruled infringing — a real, if modest, legal safety net rather than no protection at all.
Separately, and just as practically relevant: current U.S. Copyright Office guidance requires that AI-generated content be explicitly disclosed on any copyright registration application, and purely AI-generated output, without meaningful human modification, arrangement, or creative selection layered on top, cannot be copyrighted at all under current U.S. law. That has direct, immediate consequences for any business or creator relying on AI-generated visual or written content as a core commercial asset: the output itself may not be legally protectable against someone else copying it, unless genuine human creative input has been meaningfully layered into the final result — a distinction that matters enormously for exactly the kind of commercial content generation increasingly common across marketing, publishing, and design work.
The Deeper Pattern: Litigation Moving Slower Than the Technology
Step back from the specific legal mechanics, and the broader pattern connects directly to the AI-and-jobs conversation this series covered in Issue 019: the technology's actual deployment has consistently outpaced the legal and regulatory frameworks meant to govern it, across every dimension this series has examined — skill formation, employment displacement, and now, intellectual property ownership itself. Stability AI's Stable Diffusion has been in widespread commercial use for years already, generating enormous volumes of content and, in Stability's own case, facing genuine reported financial strain partly attributable to the accumulated cost and uncertainty of prolonged litigation — while the actual legal question of whether its foundational training process was lawful in the first place remains formally unresolved in every major jurisdiction that's examined it so far.
That gap between deployment and legal clarity creates genuine risk on both sides of the debate, worth naming plainly rather than picking a side. Creators whose work trained these systems have had that work commercially exploited for years while courts and legislatures work through, at a genuinely deliberate pace, whether that exploitation was lawful — a real, ongoing harm regardless of how the eventual rulings land. AI companies and the businesses now built on top of their tools face a form of genuine, unresolved legal exposure that could, depending on how the Andersen jury trial and the various appeal processes ultimately resolve, retroactively implicate years of already-deployed commercial products and the content generated through them. Neither side of this fight gets genuine certainty until the slower-moving legal process finally catches up to a technology that's already been reshaping commercial content creation for years — and based on the current trial calendar, that catch-up is still, at minimum, another year or more away.
