Over the past few months, Morten Blichfeldt Andersen has spent many hours searching OpenAI’s GPT Store. Since its launch in January, the custom bot market has been filled with a host of useful and sometimes quirky AI tools. Cartoon generators are running New Yorkerstyle illustrations and vibrant anime stills. Programming and writing assistants provide shortcuts for creating code and prose. There also is a color analysis bota identification of the spiderand a dating coach called RizzGPT. Yet Blichfeldt Andersen is only hunting one very specific type of bot: bots built without permission on his employer’s copyrighted textbooks.
Blichfeldt Andersen is publishing director at Praxis, a Danish textbook supplier. The company has embraced AI and created its own custom chatbots. But it’s currently engaged in a game of whack-a-mole at the GPT Store, and Blichfeldt Andersen is the man holding the hammer.
“I personally looked for violations and reported them,” Blichfeldt Andersen said. “They just keep coming.” He suspects that the culprits are mainly young people uploading material from textbooks to create custom bots to share with classmates – and that he has only discovered a small fraction of the infringing bots in the GPT Store. “The tip of the iceberg,” says Blichfeldt Andersen.
It’s easy to find bots on the GPT Store whose descriptions suggest they are somehow tapping copyrighted content, like Techcrunch noted in a recent article claiming that OpenAI’s store was flooded with “spam.” Using copyrighted material without permission is permitted in some contexts, but in other cases rights holders may take legal action. WIRED has found a GPT called Westeros Writer who claims to “write like George RR Martin,” creator of Game of Thrones. Another, Voice of Atwood, claims to imitate the writer Margaret Atwood. Yet another, Write Like Stephen, aims to emulate Stephen King.
When WIRED tried to trick the King bot into revealing the ‘system prompt’ that tunes its responses, the output suggested it had access to King’s memoirs Overwrite. Write Like Stephen was able to reproduce verbatim passages from the book upon request, even noting which page the material came from. (WIRED was unable to contact the bot’s developer as he did not provide an email address, phone number or external social profile.)
OpenAI spokesperson Kayla Wood says it responds to takedown requests against GPTs with copyrighted content, but declined to answer WIRED’s questions about how often it complies with such requests. She also says the company proactively looks for problem GPTs. “We use a combination of automated systems, human review, and user reporting to find and review GPTs that may violate our policies, including using third-party content without necessary permission,” says Wood.
New disputes
The GPT store’s copyright issue could add to OpenAI’s existing legal problems. The company is facing a number of high-profile copyright infringement lawsuits, including one that has been filed The New York Times and several from various groups of fiction and nonfiction authors, including big names like George RR Martin.
Chatbots offered in OpenAI’s GPT Store are based on the same technology as native ChatGPT, but are created by third-party developers for specific functions. To customize their bot, a developer can upload additional information that they can draw on to increase the knowledge baked into OpenAI’s technology. The process of accessing this additional information to respond to one’s questions is called Retrieval-Augmented Generation, or RAG. Blichfeldt Andersen believes the RAG files behind the bots in the GPT Store are a hotbed of copyrighted material uploaded without permission.