OpenAI – Application for US trademark “GPT” has failed

Original Article ↗ Hacker News Discussion ↗

OpenAI’s attempt to trademark the term “GPT” in the US has been refused on the grounds that it is a merely descriptive, widely used technical term for a class of AI models (“generative pre‑trained transformers”). Commenters largely welcome the decision as preventing a de facto land grab over a scientific acronym, comparing it to past trademark overreach around words like “gram,” “windows,” or “monster.” The exchange also explores how trademarks differ from patents, why descriptive or generic terms are hard to protect, and what this means for AI branding and for other companies using “GPT” in their product names.

Legal outcome and reasoning

  • USPTO refused OpenAI’s “GPT” trademark as merely descriptive and already widely used to denote a type of AI model (“generative pre‑trained transformer”), not a distinct brand.
  • Evidence cited: broad industry use of “GPT” generically for transformer-based models, regardless of vendor.
  • Some commenters note the office also flagged specimen issues for certain classes (e.g., not clearly showing downloadable/purchasable software in commerce).

Generic vs descriptive terms

  • Strong consensus: “GPT” describes technical characteristics and has become a generic label for a model class, so it’s legally weak as a trademark.
  • Several analogies: trying to trademark “RNA,” “polymerase chain reaction,” or “Screen Time” for usage tracking.
  • Others argue OpenAI originated “generative pre‑training” in this context, but multiple posts claim:
    • “Generative pretraining” predates OpenAI.
    • The original OpenAI paper did not use “GPT” as a name.
    • The BERT paper may have been first to label it “OpenAI GPT” (who coined the acronym is described as unclear).

Comparisons to other trademarks

  • Microsoft “Windows” is debated:
    • Some see it as precedent for generic-ish software terms being trademarkable.
    • Others emphasize it was tied to a specific OS/windowing product, not “OS” in general.
  • IBM, BMW, Apple, “App Store,” Velcro, Kleenex, LEGO, Xerox, and others are used to illustrate:
    • Descriptiveness vs distinctiveness.
    • How brands must police generic use to keep protection.
  • Trademarks are noted as class-specific; “Apple” for computers is different from “Apple” for fruit or cleaners.

Views on OpenAI’s motives and behavior

  • Many see the attempt as a land grab or “asshole move” to control a community term and threaten others using “GPT.”
  • A minority suggest a defensive rationale: better for OpenAI to be refused than for a trademark troll to succeed later.
  • Some argue OpenAI undermined its own case by using “GPT” generically and not treating it as a mark from day one.

Branding and product naming

  • Several commenters think “GPT”/“ChatGPT” are poor, technical consumer brands compared to names like Gemini/Bard.
  • Others counter that the awkward, engineer-y name actually helped it feel like real tech and is now highly recognizable, even to non‑technical users.

Broader IP and abuse concerns

  • Thread distinguishes trademarks from patents, though both are handled by USPTO and fall under “industrial property.”
  • Multiple examples of perceived trademark overreach: “Monster” (energy drink/cables), “Saga,” “Scrolls,” Entrepreneur magazine.
  • Some note prior uses of “GPT” for non-AI purposes (e.g., “Get Paid To” sites, GUID Partition Table), reinforcing that it’s not uniquely OpenAI’s term.