ChatGPT has been generating New Yorker-style cartoons that carry the signatures of real, living cartoonists who had nothing to do with them. The images are not reproductions of existing cartoons; they are new images with a named artist's signature rendered into the corner.
This is a narrower and harder problem than the usual style-imitation complaint. A model producing work that looks like an artist's is an argument about influence and training data, and reasonable people disagree about where the line falls. A model placing that artist's signature on work they did not make is an attribution claim, and it is simply false. The affected cartoonists did not sign it, were not asked, and now have images circulating with their name on them.
It also lands badly next to the week's other OpenAI disputes - a rejected benchmark figure and a contested set of firings - because all three turn on whether a published claim is accurate. A signature is the most compact claim a piece of work can carry, and it is one the model should be least willing to invent.
Sources
Futurism · TechSpot · Creative Bloq