Nieman Lab documented more than 15 cartoonists whose signatures appeared on AI-generated images without permission, raising sharp questions about attribution, identity and training data
Signed, But Not By Them
When a fan asked ChatGPT for a New Yorker-style cartoon, the result was a single-panel drawing featuring Dolly Parton and Tim Curry in heaven. In the corner was a signature: BLOPER, the pen name of the New Yorker cartoonist Brendan Loper. Loper had nothing to do with it. That cartoon, which circulated widely online, was the starting point for an investigation by Nieman Lab reporter Andrew Deck, published on October 5 under the headline ChatGPT is adding real cartoonists’ signatures to fake New Yorker cartoons.
Deck documented more than 15 cartoonists whose signatures appeared on AI-generated images without permission. Among those named were Brendan Loper, Harry Bliss, Emily Flake, Joe Dator, Pat Byrnes, Peter Vey, Jason Adam Katzenstein, George Booth, Liza Donnelly, Ellis Rosen, Saul Steinberg and Warren Miller. Some are working cartoonists. Others, including Steinberg, are long dead, their signatures now reappearing on drawings they never made.
How It Came To Light
Loper first learned of the problem in May, when a stranger alerted him on Instagram. He later found more examples in a Reddit forum dedicated to ChatGPT. Deck confirmed the pattern through his own tests. The signatures were not occasional glitches. They appeared repeatedly when users asked for cartoons in the New Yorker style.
The cartoonists’ reactions were striking. Loper described the experience as feeling like a violation of his personhood. Katzenstein called the output slop that is pretending to be him. Flake said the work looked like a distant cousin of hers. Dator compared it unfavourably to credit card fraud.
The Companies Respond
An OpenAI spokesperson told Nieman Lab that the company focuses on tools that empower human creativity and thanked the community for flagging bugs. After Deck raised the issue, ChatGPT began showing a guardrail warning for some prompts, though it still signed some cartoons. OpenAI declined to say how its models might have learned the signatures.
A spokesperson for the New Yorker said that Conde Nast, its parent company, has never allowed a large language model developer to train on its cartoons, and that reproducing the New Yorker’s logo alongside cartoons is not permitted under any such deal. That statement raises an obvious question: if the cartoons were not licensed for training, how did the model learn to reproduce the cartoonists’ signatures so precisely?
Why A Signature Matters
For a cartoonist, a signature is not decoration. It is identity, authorship and professional reputation compressed into a few strokes. It tells readers who made the work and who stands behind it. A fake signature on an AI-generated image does more than infringe on style. It attributes words and images to a real person who did not create them, and potentially associates that person with jokes, opinions or quality they would never endorse.
This is a problem that journalism should recognize immediately. A byline serves the same function as a cartoonist’s signature. It is a promise of accountability. If AI systems can attach real journalists’ bylines to synthetic articles, or real photographers’ credits to generated images, the basic infrastructure of trust that journalism depends on begins to erode.
A Wider Pattern
The cartoon case is part of a broader set of concerns about how generative AI systems treat the work and identities of creators. Publishers, authors, artists and photographers have filed numerous lawsuits arguing that their work was used to train AI models without permission or payment. The question of attribution, whether AI outputs credit, misattribute or fabricate sources, has also become a recurring issue in AI-generated news summaries and search answers, where errors in sourcing can mislead readers.
The cultural conversation around AI and authenticity has taken many forms, including comedy. The London Prat recently imagined Piers Morgan demanding emergency action after a chatbot surpassed human politeness a comic take on the anxieties AI provokes among media figures. The cartoonists’ complaint is less comic. Their concern is not that the machine is too polite, but that it is putting their names on work they did not make.
What Should Change
Several practical steps could reduce the harm. AI companies could prevent their systems from reproducing identifiable signatures, logos and bylines, and make those safeguards robust rather than partial. They could disclose more about the sources of their training data, so creators can understand how their work was used. Publishers and creators could secure clearer contractual protections in any licensing deals. And regulators could consider whether false attribution of authorship to real people should carry specific legal consequences.
There is also a question of public trust. Readers who encounter a cartoon signed with a familiar name will reasonably assume the named artist made it. When that assumption is false, the deception is not only against the cartoonist but against the audience.
The Human Behind The Line
New Yorker cartoons are among the most distinctive forms of visual journalism in American media, produced by a small community of artists who spend years developing their voices. Each cartoon carries the mark of a particular sensibility. The appearance of their signatures on machine-made imitations is a reminder that AI systems are built from human work, and that the people who did that work have a legitimate interest in how it is used and credited. As Deck’s reporting shows, that interest is not yet adequately protected.