Potential Trademark Liability from GenAI

By Wei Wei Jeang

In a case involving The New York Times and Perplexity AI, The Times has raised concerns that hallucinatory content generated by the AI tool has been falsely attributed to the publication. The Times alleged copyright and trademark infringement in the U.S. District Court for the Southern District of New York

While much of the litigation centers on copyright infringement, The Times also argued that Perplexity’s fabricated information, or “hallucinations,” and falsely attributed them to the Times by displaying the newspaper’s registered trademarks alongside the material. The Times alleges that this is a false designation of origin of the hallucinogenic GenAI output to the publication that is likely to confuse users and tarnish the newspaper’s reputation for accuracy.

The New Legal Frontier

Traditional trademark infringement is based on human conduct: the infringer intentionally uses a trademark in promoting its goods or services with or without knowledge that the mark has a prior trademark user. GenAI disrupts that framework in at least two important ways.

1. No Traditional “Intent” to Confuse

In classic infringement cases, intent is one of the factors analyzed by the court. Courts often look at whether the defendant adopted a mark with knowledge of the plaintiff’s brand — and at times treat bad intent as evidence of likely confusion.

With GenAI systems, even though the GenAI tool creator trained its large language model (LLM) using data that includes The Times's content:

• The GenAI tool provider does not directly participate in creating the output to a user prompt.

• The LLM generates responses probabilistically.

• The allegedly infringing act may not be foreseeable.

  
  

2. The Output Is Dynamic and User-Specific

Traditional infringement involves a static representation — a label, website, advertisement, or product packaging that is consistently displayed to consumers. GenAI output, by contrast:

• Is generated in real time in response to user input

• Can vary with each user prompt

• May not be stored or publicly visible

• Can change depending on variations in how the prompt is phrased

This lawsuit is the latest in a series of more than 50 copyright-related suits filed in the United States against AI tech companies. In September 2025, Anthropic agreed to pay at least $1.5 billion to settle Bartz v. Anthropic, in which authors alleged that the AI company trained its models on hundreds of thousands of pirated books, in what the plaintiffs called “the largest publicly reported copyright recovery in history.” In October 2025, Reddit filed a suit against Perplexity for allegedly bypassing its security measures to scrape user-generated content, which Perplexity had admitted was a “top-tier source” for its data.