1. Getty Images v. Stability AI: Output-based trademark liability in the UK
The UK High Court’s November 2025 decision in Getty Images v. Stability AI marked the first major UK ruling on how trademark law applies to generative AI systems.
Although Getty failed on its central copyright claims due to territorial limitations, it achieved partial success on trademark infringement, and that aspect of the ruling may have the longest-lasting impact.
Watermarks as trademarks
The court accepted evidence that earlier versions of Stable Diffusion could generate images containing Getty and iStock watermarks under realistic prompting conditions. The court held that:
If AI-generated outputs reproduce trademarks in the course of trade, liability may arise.
Watermark replication can confuse consumers as to origin, endorsement, or licensing.
However, the court limited liability to historic and isolated instances, noting that Stability’s filtering measures and hosted platforms substantially reduced the risk over time.
Why this case matters
This decision draws a crucial distinction:
Training data and model weights were not deemed infringing copies.
Outputs, however, remain squarely within trademark law’s reach.
For brand owners, the ruling confirms that trademark enforcement may be more viable than copyright claims where AI outputs replicate logos, watermarks, or brand identifiers.
For AI developers, the case emphasizes the importance of output governance, including watermark detection, filtering, and documented mitigation efforts.
2. The New York Times v. Perplexity AI: Hallucinations as trademark harm
The most consequential trademark dispute of 2025 emerged from an already crowded copyright battlefield. In December 2025, The New York Times Company sued Perplexity AI in the Southern District of New York, alleging both copyright and trademark infringement arising from Perplexity’s AI-powered “answer engine.”
While much public attention focused on the Times’ extensive copyright allegations, the trademark claims may prove just as influential.
Trademark claims based on AI “hallucinations”
The Times alleged that Perplexity’s system not only reproduced its articles but also generated fabricated information falsely attributed to the Times, displaying the newspaper’s registered trademarks alongside inaccurate or incomplete content. According to the complaint, this amounted to:
False designation of origin under the Lanham Act
Trademark infringement and dilution, based on reputational harm
Misleading, incomplete reproductions of Times content without disclosure
This theory reframes AI hallucinations as a trademark problem, not merely a quality-control issue. When an AI system attributes false content to a trusted news brand, the risk is not just confusion, but erosion of goodwill built on accuracy.
Why this case matters
If courts accept this theory, AI companies could face trademark liability even when outputs are not verbatim reproductions. The key question becomes whether brand signals embedded in AI interfaces (logos, source labels, attribution links) create consumer confusion when the underlying content is unreliable.
3. iyO, Inc. v. IO Products: Trademark risk before product launch
Not all of 2025’s important AI trademark cases involved AI outputs. In iyO, Inc. v. IO Products, Inc., the Ninth Circuit addressed a more traditional, but increasingly common, problem in the AI sector: brand conflict at the announcement stage.
The court affirmed a temporary restraining order preventing IO Products, a hardware startup acquired by OpenAI, from using the “IO” mark in connection with products related to iyO’s AI-based “audio computer.” iyO owns the registered IYO mark and brought suit shortly after IO’s highly publicized May 2025 launch announcement.
IO argued that the dispute was premature because it had not yet released or sold a product. The Ninth Circuit disagreed, reiterating that trademark infringement does not require actual sales. The court held that IO’s launch video, designed to generate anticipation for a forthcoming AI device, could itself constitute infringing advertising. Because IO had a working prototype, planned to compete directly with iyO, and intended to market its product under the challenged mark in 2026, the alleged infringement was deemed “sufficiently imminent.”
Why this case matters
The Ninth Circuit also upheld the finding of irreparable harm, crediting evidence that IO’s branding announcement had already disrupted iyO’s fundraising efforts and threatened its brand identity.
Although the injunction was narrowly limited to products sufficiently similar to iyO’s, the decision reinforces a broader point echoed throughout 2025’s AI trademark cases: in fast-moving AI markets, branding decisions, and missteps, can create actionable trademark risk well before a product ever reaches consumers.
Takeaway
Looking ahead, trademark law will increasingly shape how AI companies innovate and compete. We are entering a world where brand identity and consumer trust are not just marketing concerns: they are legal considerations that must be built into AI design, output controls, and launch strategies from day one.
As AI systems become more capable of generating content, reproducing brand signals, or even creating new brand-like identifiers, trademark practitioners will need to think proactively about monitoring, enforcement, and risk management in ways that were unimaginable a few years ago. The cases of 2025 are just the beginning; the next decade will test how flexible trademark law can be in keeping pace with rapidly evolving technology.
