OpenAI, the startup behind the wild chatbot ChatGPT, has recently felt the need to protect its brand, and it’s easy to see why. Numerous outfits, such as ThreatGPT, MedicalGPT, DateGPT, and DirtyGPT, have filed trademark applications with the United States Patent and Trademark Office in recent months.
These companies are taking advantage of the incredible popularity of ChatGPT, which was launched by OpenAI in November and is built on the company’s deep learning model, GPT-4, released just last month.
Unsurprisingly, OpenAI applied for a trademark for “GPT” in late December, which stands for “Generative Pre-trained Transformer.” Last month, OpenAI even petitioned the USPTO to expedite the process, citing the increasing number of “myriad infringements and counterfeit apps” that are beginning to emerge.
OpenAI faced a setback last week when its petition to expedite the trademark application process was dismissed.
The United States Patent and Trademark Office stated that OpenAI’s lawyers failed to pay the required fee and provide adequate documentary evidence to support their request for special action.
As a result, the decision on the trademark application could take up to five more months, according to Jefferson Scher, a partner in the intellectual property group of Carr & Ferrell and chair of the firm’s trademark practice group. Even then, the outcome is uncertain, Scher explains.
Despite this uncertainty, Scher believes OpenAI has a good chance of securing the patent. For example, he points out that IBM, short for International Business Machines, is a brand with a descriptive origin, even though the description is weak. Scher acknowledges that having a descriptive origin is no guarantee that OpenAI will be able to obtain the GPT trademark, but such precedents can be helpful.
Scher also addresses whether OpenAI might encounter resistance because “T” in GPT stands for “Transformer,” which is the name of a neural network architecture developed by Google in 2017 that is widely used. He believes GPT can still be a brand, even with a descriptive origin.
According to Scher, another factor that could benefit OpenAI is the company’s prior use of “GPT.” OpenAI has been using the acronym for years, with the release of its original Generative Pre-trained Transformer model, GPT-1, in October 2018.
However, Scher notes that this is a unique situation. Typically, a brand has gradually built its market presence when claiming a trademark. In OpenAI’s case, the company was primarily known to AI researchers until last year when it released DALL-E 2, a mesmerizing deep learning model that generates digital images, followed by ChatGPT, which turned the company into an overnight sensation.
Even if the USPTO examiner approves OpenAI’s application, the process will move to an opposition period, during which other market participants can argue why the agency should deny the “GPT” trademark. Scher explains that OpenAI must establish that “GPT” is proprietary and that the public perceives it as such rather than as a reference to generative AI more broadly.
Establishing public perception regarding the “GPT” trademark can be complex for the USPTO. One option is to survey randomly selected Americans, but this would require a significant budget that the government is unlikely to cover. Therefore, any challenger to OpenAI’s trademark must finance such an endeavor.
Another approach to determining public perception is to evaluate how “GPT” has been used in various public domains, such as late-night talk shows and public writing. If the public does not recognize it as proprietary, a trademark trial would determine whether it is protectable. Such a legal process could be lengthy, and OpenAI would understandably prefer to avoid it.
The question arises as to why OpenAI did not take steps to protect the “GPT” trademark earlier. According to Scher, the company was likely caught off guard by its rapid success.
OpenAI is reportedly attempting to register a related trademark in China, where it has not yet launched ChatGPT and may not be able to do so. Scher believes that “we’ve crossed a line where GPT is not three random letters” and that it is unsafe for startups to adopt it.
Scher suggests that OpenAI may benefit from a facet of trademark law that recognizes the fame of a registered mark as a dominant factor. The company can prevent its general use if OpenAI can establish that “GPT” is a famous trademark. However, this would be expensive, and chasing after offenders would require significant resources.
This could be one positive outcome for the company in this lengthy process. As OpenAI gains more users and coverage, it becomes more likely that it will be able to establish itself as a famous trademark, even outside of its sphere.
When asked whether OpenAI is known to the average person or household, Scher responded that they might be closing in on that level of recognition.