ChatGPT is adding real cartoonists' signatures to fake New Yorker cartoons
ChatGPT’s image generator has been producing New Yorker-style cartoons that include what appear to be real cartoonists’ signatures, raising concerns about copyright, trademark infringement, and outright forgery. Commenters debate whether liability should fall on users or AI companies, and whether current laws on intellectual property, impersonation and “right of publicity” are equipped to handle such cases. The incident is also used to highlight broader worries about generative AI’s reliance on unlicensed training data, its impact on creative professions, and the limits of current safeguards.
Nature of the signature problem
- Many commenters say the model is just reproducing common pixel patterns: New Yorker-style cartoons in the training set almost always have an artist signature, so the model statistically “expects” a signature in that corner.
- Similar artifacts (Getty watermarks, logos, Japanese hanko seals) are cited as parallel examples.
- Several argue this is a clear product bug and should be trained or filtered out; others say it’s just a predictable property of diffusion-style generators.
Legal characterization and liability
- Debate over whether this is mainly copyright infringement, trademark infringement, forgery, or “right of publicity” / impersonation.
- Some argue reproducing a stylized signature is more serious than copying style, because it falsely attributes authorship and risks reputational harm.
- Disagreement on who is liable:
- One camp says the user who prompts and publishes is responsible, like misusing Photoshop.
- Another says the AI provider is selling a service that forges signatures at scale and should face lawsuits or regulatory action.
- Some think this is strong lawsuit material and may be what finally forces legal clarity.
Ethics of training data and “plagiarism as a service”
- Many see this as evidence that models are effectively trained on “stolen” cartoons and other copyrighted works, then used to create market-substituting fakes.
- “Plagiarism as a service” / “copyright laundering” / “forgery as a service” are recurring characterizations.
- Others counter that everything is derivative, humans also learn by copying, and existing fair-use case law for search/sampling might justify training.
Broader AI concerns (jobs, harms, externalities)
- Thread branches into wider critiques: massive copyright violations, job displacement (especially creative work and junior coders), energy use, and concentration of wealth.
- Some participants argue AI is a historic “theft of labor”; others emphasize real productivity gains in coding, math, and research and want a split between “practical” and “artistic” applications.
Proposed mitigations
- Suggestions include:
- Default removal of signatures/watermarks via preprocessing or a second AI pass.
- Stronger output filters and negative prompts.
- Watermarking AI output (invisible rather than visible “AI signatures”).
- Ethically sourced or opt-in datasets, though feasibility is disputed.
- Creators registering signatures as trademarks and pursuing collective legal action.