Loading page…
A warning did not stop some signed outputs. The New Yorker says its OpenAI deal did not authorize cartoon training, and image verification cannot establish an artist’s approval.
Listen to this story
Nieman Lab found that ChatGPT-generated cartoons carried signatures associated with more than 15 New Yorker contributors, creating a risk that readers will mistake generated images for artists’ work. ChatGPT began warning users that requests for New Yorker-style cartoons could violate its rules, but the warning did not prevent every output from bearing a real artist’s name. OpenAI has not explained how the signatures appeared; the findings also distinguish image-origin checks from proof that a named artist made or approved an image.
Cartoonist Brendan Loper said he had been alerted to another ChatGPT image bearing his pen name in May.
Emily Flake said some images combined several artists’ styles while displaying a signature recognizably hers.
Condé Nast’s 2024 content deal with OpenAI did not authorize training on New Yorker cartoons or using its art and logo in image generation, the publication said.
Brendan Loper began hearing from people who thought he had drawn a cartoon circulating online. He hadn’t—but his pen name was in the corner. Nieman Lab found signatures associated with more than 15 New Yorker contributors on ChatGPT-generated cartoons, according to Digital Trends’ October 8 coverage. The problem extends beyond imitation to mistaken authorship.
The circulating image carried “BLOPER,” Loper’s pen name, even though he had neither drawn nor signed it. This was not his first encounter with the behavior: he told Nieman Lab that someone had alerted him to another ChatGPT-generated image bearing his name in May, Futurism reported on October 7.
Other affected cartoonists included Harry Bliss, Emily Flake, Joe Dator, Pat Byrnes, Peter Vey and Jason Adam Katzenstein. Flake described drawings bearing her pen name as a blend of several artists’ styles, while the signature was recognizably hers.
For Loper, a signature is a deliberate claim of authorship. He told Nieman Lab that he puts his name on work he wants recognized as coming from his hand.
After Nieman Lab raised the issue, ChatGPT began warning that requests for New Yorker-style cartoons could conflict with its rules on similarity to third-party content. But Nieman Lab still generated some cartoons carrying real artists’ names after the warning appeared.
OpenAI said it appreciates reports of bugs and unintended model behavior so it can address them. It declined to explain how the artists’ names appeared in the outputs.
Condé Nast signed a content deal with OpenAI in 2024 that included The New Yorker among its brands. But a New Yorker spokesperson said the agreement did not authorize training on cartoons or using its logo and art in image generation, Mashable reported. The publication also said contributing cartoonists retain copyright in their work.
OpenAI’s public image verification tool answers a different question from the one facing readers of a signed cartoon. It checks for signals associated with images generated by OpenAI products. It cannot confirm that the artist named on an image made or approved it.
Nor does an absent signal prove human authorship. Sharing or editing can strip metadata—the information attached to an image file—or make a watermark harder to detect. Digital Trends recommends checking an artist’s portfolio or the publication’s archive before crediting a familiar signature.
Loading discussion...
Join the conversation
Explain whether a false credit crosses a different line.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.