ChatGPT Put Cartoonists’ Names on Images They Didn’t Draw

A generated image can arrive with an artist’s signature already attached. For a marketing team, that small mark can turn a usable-looking creative file into a false claim about who made it.

Nieman Journalism Lab reported on October 5 that ChatGPT had produced New Yorker-style cartoons bearing signatures associated with the magazine’s cartoonists. The examples had appeared online over the summer, including one posted on August 26. The Lab also tested the behavior itself. It found signatures linked to several contributors, including Brendan Loper and Emily Flake. Some generations had no signature, while others included marks that didn’t correspond to a known name.

That makes this a recurring output to check, rather than a newly announced feature. Nieman Lab also found outputs that reproduced The New Yorker’s logo. A Condé Nast spokesperson told the Lab that the publisher had not given an AI developer permission to train on its cartoons. OpenAI gave the Lab a general statement describing the behavior as unintended and saying it welcomed reports of bugs. After the reporter raised the issue, a new warning appeared for some prompts that requested New Yorker-style cartoons. Nieman Lab said signatures still appeared in later tests as of its October 5 publication. It remains unclear whether that behavior has since been fixed. OpenAI declined to explain how the signatures may have entered the model’s output.

The distinction matters for anyone using generative images in a campaign. A signature can suggest authorship, endorsement or a paid collaboration. When a marketer publishes an image without checking the mark, audiences may reasonably believe the named artist created or approved it. The same risk can arise with a creator’s handle, a photographer’s watermark or a brand logo, even when the team asked only for a general visual idea.

The artists described the personal cost. Flake told Nieman Lab, “It’s like somebody attributed a quote to me that I didn’t say.” Loper told Nieman Lab, “my name is my name.” Both comments point to attribution as a part of the creative, not a decorative detail that can be ignored after generation.

OpenAI’s September system card for ChatGPT Images 2.5 says the models were trained on a mix of public internet information, third-party material and content from users or human trainers. It describes filtering and image safety checks for harmful or sensitive outputs. The card doesn’t specify whether the system evaluates generated signatures or verifies who authored a depicted work. It also doesn’t identify whether any particular cartoon archive appeared in the training data.

That gap is useful for marketers to understand. A model card can describe broad data sources and safety controls while leaving a practical workflow question unanswered: will the tool catch a false byline before the image reaches a campaign review? The public documents cited here don’t answer that question.

The first control belongs in the approval process. Review the image itself at full size, including corners, borders, labels and background details. Look for signatures, watermarks, handles, logos and recognizable characters. Compare them with the creative brief and the asset’s source record. If the mark is not intended and authorized, remove the asset or correct it before publication.

Teams can also separate concept exploration from production approval. A generated sketch may help a designer explore composition. A final paid or owned asset needs an origin record, rights review and a named approver. A prompt asking for “no signature” can reduce the chance of a mark appearing, but the rendered file still needs inspection.

This story is adjacent to our coverage of consumer trust in AI product imagery and OpenAI’s visual ad test in ChatGPT. Those stories cover how AI-generated or AI-served images reach consumers. Here the concern is whether a tool assigns an image to a person who didn’t make it. For brand teams, the question is whether the image falsely signals that a specific person made or endorsed it.

Marketers have spent years building creator partnerships around recognizable voices and visual styles. Generative tools make it faster to explore those styles, and that speed raises the value of provenance checks. The signature at the edge of the frame may be the detail that changes what the image claims.