# What Did ChatGPT Actually Reproduce?

According to Nieman Lab, ChatGPT generated cartoons in the visual idiom of The New Yorker and included the signatures of real, working cartoonists on the output. Nieman Lab's October 5, 2026 report frames this as a signature problem, not a style problem — the model is not merely mimicking a house aesthetic, it is attaching a specific human's mark of authorship to images that human did not make. That distinction is the whole story. Style imitation has been litigated for two years and remains contested. Signature reproduction is narrower and more concrete: a signature is a personal identifier, closer to a trademark or a right-of-publicity asset than to a copyrighted drawing. The New Yorker's own cartoon archive, which maintains signed originals going back decades, is the reference corpus that makes the forgery legible — a reader recognizes the signature because the publication trained them to.

# Why Do Image Models Learn Signatures At All?

Signatures in cartoon art are not metadata. They are pixels. Every signed New Yorker cartoon in a training set teaches the model that a small, high-contrast cursive mark in a corner is part of the composition. The model has no concept of "attribution" — it has a statistical association between a visual style and a visual mark.

That is why this is not a simple filter fix. Removing signatures from training data would degrade the model's ability to render any hand-drawn aesthetic, because the signature is spatially entangled with the drawing. The more tractable fix is post-generation detection, which means OpenAI would need a signature classifier running on every image output — an expensive, latency-adding step the company has so far avoided.

# Is This a Copyright Problem or a Forgery Problem?

It is closer to forgery. Copyright protects the drawing; a forged signature misrepresents who made it. The New Yorker has not commented publicly on the Nieman Lab report as of this writing, and OpenAI has not issued a statement specific to signature reproduction. But the legal exposure runs through right-of-publicity statutes in states like California and New York, which protect a person's name and likeness from commercial misappropriation — a doctrine that does not require the plaintiff to own a copyright in the underlying work.

Approach	What It Protects	Enforcement Cost	Applies to Signatures?
Copyright (drawing)	The specific cartoon image	High — requires registration and substantial similarity	No
Right of publicity	Name, likeness, signature as identifier	Medium — state-by-state	Yes
Trademark (publication mark)	The New Yorker's brand identity	Medium — requires consumer confusion	Partially
Content Credentials (C2PA)	Provenance of the file	Low — technical standard	Yes, if adopted
Verdict	Right of publicity is the strongest near-term lever; Content Credentials is the strongest long-term fix.

# Who Actually Bears the Cost?

The cartoonists. A forged signature on an AI image does not damage OpenAI's reputation with its paying customers — most will never see the output. It damages the named cartoonist, whose signature now appears on work they did not create and cannot disavow at scale. The New Yorker, as the publication whose visual identity is being cloned, has a brand interest but not a personal one. OpenAI, per its own usage policies, prohibits impersonation, but enforcement is reactive. The asymmetry is structural: one model output can forge a signature that took a cartoonist decades to establish.

# What Changes in Editorial Workflow?

The practical response is provenance infrastructure. Adobe's Content Credentials, built on the C2PA standard, attaches a cryptographically signed manifest to an image at creation. If The New Yorker required C2PA credentials on submitted illustration, a forged signature would be detectable at intake. Nieman Lab's report does not mention whether The New Yorker has adopted such a requirement, and there is no public evidence it has.

# Predictions

1. By Q2 2027, OpenAI will ship a signature-suppression filter for DALL·E and GPT image outputs, and independent researchers will demonstrate a bypass using prompt-level obfuscation within 60 days of release.
2. By Q4 2027, at least one U.S. state attorney general will open a right-of-publicity inquiry into generative models reproducing personal signatures, with California the most likely jurisdiction given its statute and the concentration of affected artists.
3. By mid-2027, The New Yorker or its parent Condé Nast will publicly require C2PA Content Credentials on submitted illustration, making it the first major U.S. magazine to do so.

# Article Summary

• The signature reproduction is a training-data artifact, not a hallucination — signatures are pixels the model learned as part of the cartoon composition.
• Right of publicity, not copyright, is the strongest legal lever because it protects a name independent of any copyrighted work.
• The reputational cost falls on individual cartoonists, not OpenAI or The New Yorker, which is why the market response will be contractual rather than litigious.
• Provenance standards (C2PA) are the only scalable fix, which shifts value to Adobe and verification vendors.
• Any signature filter OpenAI ships will be a speed bump, not a wall — expect bypasses within weeks.