Generative AI has moved faster than the law meant to govern it. Most of the resulting litigation so far has focused on copyright who owns the training data, and whether scraping it was lawful.
But a 2025 lawsuit is pulling a second, quieter body of law into the spotlight: trademark dilution. And for brand owners, AI developers, and businesses building on generative tools, the implications reach well beyond one courtroom.
The Case: Getty Images (US), Inc. v. Stability AI, Ltd.
In August 2025, Getty Images filed suit against Stability AI — the company behind the popular Stable Diffusion image-generation model — in the U.S. District Court for the Northern District of California (Getty Images (US), Inc. v. Stability AI, Ltd., 3:25-cv-06891, N.D. Cal.).
Getty’s core allegation: Stability AI trained Stable Diffusion on millions of Getty photographs and their metadata without authorisation. Alongside the expected copyright claims, Getty brought three claims under the Lanham Act — the primary U.S. federal trademark statute:
• Trademark infringement
• False designation of origin
• Trademark dilution
The trademark angle stemmed from a specific, recognisable problem: Stable Diffusion was generating images that partially reproduced the Getty Images watermark. Even when incomplete or distorted, Getty argued, these fragments were enough to mislead viewers into believing the AI output was somehow connected to, endorsed by, or sourced from Getty.
In April 2026, the court denied Stability AI’s motion to dismiss, holding that Getty had plausibly alleged all three trademark claims — enough to proceed to discovery. It is important to be precise about what this means: the court did not rule that Stability AI is liable.
It ruled that Getty’s claims are strong enough, on paper, to be tested through the litigation process. Still, the reasoning behind that ruling matters enormously for how trademark law will treat AI-generated content going forward.
Why This Case Matters Beyond Getty and Stability AI
1. AI-generated images can be “goods” under the Lanham Act
Stability AI argued that AI-generated images shouldn’t count as “goods” for trademark purposes at all — essentially, that the statute wasn’t built for algorithmic output. The court rejected this. It held that commercially distributed AI-generated images can constitute goods within the meaning of the Lanham Act.
This is a critical point for any business monetising generative AI outputs. It confirms that trademark liability can attach to the output of an AI system, independent of whatever copyright issues exist in the training data. In other words, even if a company resolves its copyright exposure, a mark-mimicking output can still create a live trademark problem.
2. Algorithmic origin doesn’t change the confusion analysis
The court’s reasoning treated AI-generated images the same way it would treat any conventional commercial product bearing a misleading mark. The fact that an image was produced by a diffusion model rather than a human designer did not change the legal inquiry. If the output is capable of misleading consumers about its commercial source, sponsorship, or affiliation, ordinary trademark principles apply.
This closes off a defence that many AI developers have been counting on that the “black box” nature of generative output somehow insulates it from consumer-protection law. It doesn’t.
3. False designation of origin claims can survive at the pleading stage
The court also allowed Getty’s false designation of origin claim to proceed. This Lanham Act provision targets misleading representations about the source, sponsorship, affiliation, or approval of goods and services a broader net than infringement alone, and one that AI companies distributing branded-looking outputs should watch closely.
Trademark Dilution: Why It Doesn’t Require Confusion
This is the part of the ruling with the widest long-term implications. Trademark infringement requires a likelihood of consumer confusion. Trademark dilution does not.
Dilution protects famous marks from two distinct harms:
• Blurring — uses that weaken a famous mark’s distinctiveness, even absent confusion.
• Tarnishment — uses that damage a famous mark’s reputation, typically by association with something unsavoury or low-quality.
Getty argued its watermark qualifies as a famous mark, citing decades of nationwide use, wide media recognition, and a large global customer base. At the pleading stage, the court found these allegations sufficient to let the dilution claim proceed.
The practical significance: even if every user of Stable Diffusion fully understood they were looking at AI-generated output no confusion whatsoever the repeated, degraded reproduction of the Getty watermark could still erode what makes that mark distinctive. A famous mark can be harmed by sheer overexposure in low-quality, unauthorised contexts, regardless of whether anyone believes Getty actually produced the image.
For brand owners with globally recognised marks logos, watermarks, stylised names this is a meaningful new front. It suggests generative AI systems that tend to reproduce fragments of famous marks in their outputs may create exposure that has nothing to do with whether users are fooled.
What the Ruling Does Not Decide
It’s worth being precise about the limits of the April 2026 order:
• It is not a final judgment on liability.
• It does not establish that Stability AI infringed or diluted Getty’s marks.
• It only holds that Getty’s allegations, taken as true at this stage, are legally sufficient to proceed to discovery.
The case now moves into the evidentiary phase, where Getty will need to substantiate its claims with proof — how often the watermark artefacts appear, how recognisable they are, and what actual or likely effect they have on consumers and on the Getty brand.
What This Means for Businesses, Brand Owners, and AI Developers
For brand owners and trademark holders
For businesses building or deploying generative AI tools
• Liability can arise from the output of a model, not just from how it was trained. Cleaning up your training data does not automatically clean up your legal exposure.
• “It’s just an algorithm” is not a viable defence to trademark claims. Courts are applying conventional trademark principles to AI-generated commercial content.
• Commercially distributing AI outputs likely qualifies you as dealing in “goods” under the Lanham Act, opening the door to infringement, false designation of origin, and dilution exposure.
For companies using AI-generated content in marketing or products
• Outputs that inadvertently carry watermark fragments, stylistic signatures, or other source-identifying elements of third-party brands can create downstream liability for the business using them not just the AI developer.
• Due diligence on the provenance and cleanliness of AI-generated assets is now a trademark risk-management issue, not just a copyright one.
Key Takeaways
• Trademark law not just copyright is emerging as a serious front in generative AI litigation.
• AI-generated images can constitute “goods” under the Lanham Act when commercially distributed.
• Algorithmic origin does not exempt AI outputs from ordinary trademark confusion analysis.
• Dilution claims do not require consumer confusion repeated, unauthorised reproduction of a famous mark’s elements can erode its distinctiveness or damage its reputation on its own.
• The Getty v. Stability AI ruling is a pleading-stage decision, not a final verdict but it signals that courts are willing to apply established trademark doctrine to novel AI fact patterns.
Need help assessing your brand’s exposure to AI-related trademark risk, or strengthening your trademark portfolio’s enforcement position? Legacy Partners’ IP and Trademark Advisory team can help you. info@legacypartners.global
