Lawyer Says Studio Ghibli Could Take Legal Action Against OpenAI

What the headline actually reported

The headline did not mean that Studio Ghibli had sued OpenAI. It referred to a legal-risk assessment published by Futurism on March 28, 2025, which quoted Rob Rosenberg, a former Showtime general counsel and commentator on artificial-intelligence law. Rosenberg said Studio Ghibli might have grounds to pursue OpenAI, but he also emphasized that the question was legally complex and that existing precedent did not clearly resolve the relevant generative-AI disputes.

#	Preview	Product	Price
1		Hayao Miyazaki	$31.49	Buy on Amazon
2		The Art of Spirited Away	$28.00	Buy on Amazon
3		The Art of Howl's Moving Castle	$16.59	Buy on Amazon
4		The Art of My Neighbor Totoro: A Film by Hayao Miyazaki	$15.53	Buy on Amazon
5		The Art of Princess Mononoke	$18.59	Buy on Amazon

That distinction matters. The available reporting did not establish a filed lawsuit, a court finding of infringement, a settlement, or OpenAI liability. The defensible description is that lawyers identified several possible claims arising from the viral spread of Studio Ghibli-style images made with ChatGPT.

How the controversy began

On March 25, 2025, OpenAI announced native image generation in GPT-4o. The launch materials described a system that could create images from prompts, transform uploaded photographs, follow detailed visual instructions, and work across a wide range of styles.

#1 Best Overall

Hayao Miyazaki

$31.49

Buy on Amazon

Within roughly a day, users began uploading photographs and asking ChatGPT to recreate them in imagery associated with Studio Ghibli. The resulting trend became widely known as “Ghiblification.” It spread rapidly across social media and generated enough demand that OpenAI temporarily applied rate limits during the initial surge.

The episode also received extra attention after a Studio Ghibli-style social-media image associated with OpenAI chief executive Sam Altman circulated publicly. Those facts help explain the commercial and publicity context, but they do not, by themselves, prove that Studio Ghibli endorsed the tool, licensed its works, or suffered legally actionable infringement.

OpenAI said its system was designed to refuse requests to imitate living individual artists while permitting broader studio styles. That was a product-policy distinction, not a ruling about what copyright or trademark law permits. A studio style and the work of a living individual artist can also raise different legal and factual questions.

What was established—and what was not

The possible Lanham Act theory: false association or endorsement

Rosenberg’s principal theory was not simply that an image looked like a Ghibli film. He suggested that Studio Ghibli could potentially rely on the U.S. Lanham Act, including theories involving false advertising, trademark infringement, or unfair competition.

Section 43(a) of the Lanham Act, 15 U.S.C. § 1125(a), can impose civil liability when a representation used in commerce is likely to confuse consumers about affiliation, connection, association, origin, sponsorship, or approval. It also addresses certain misleading commercial descriptions and representations.

Rank #2

Sale

The Art of Spirited Away

• New
• Mint Condition
• Dispatch same day for order received before 12 noon
• Guaranteed packaging
• No quibbles returns

$28.00

Buy on Amazon

Applied to this controversy, the question would be whether OpenAI—or another commercial promoter—presented the capability in a way that made consumers think Studio Ghibli had authorized, sponsored, or partnered with it. A promotional message explicitly inviting customers to make “Ghibli-style” images could create a different legal question from a user’s private prompt that merely refers to a visual influence.

Several facts would be important:

• Whether OpenAI used “Studio Ghibli,” related marks, or recognizable branding in advertising or product presentation;
• Whether the reference was a nominative description, such as identifying the kind of result users were seeking, or was presented as branding for the feature itself;
• Whether consumers were likely to believe Ghibli had approved, licensed, or endorsed the capability;
• Whether the challenged material appeared in commercial advertising, a user-generated post, or an editorial or expressive context; and
• Whether any use of Ghibli marks was likely to damage or trade on the studio’s established goodwill.

That is why the legal theory was described as possible rather than certain. A name reference in a prompt does not automatically create trademark liability, and a viral user post does not automatically become an OpenAI advertisement. The presentation, context, audience, and likelihood of confusion would matter.

Training-data claims are separate from output claims

The controversy involved two different copyright questions that are often merged in online discussion.

1. Was copyrighted material used to train the model?

AP reported that attorney Josh Weigensberg questioned whether the model had been trained on works by Hayao Miyazaki or Studio Ghibli and whether OpenAI had permission or a license. If copyrighted works were copied or processed for training without authorization, that could create a separate legal dispute involving reproduction, licensing, market harm, and possible defenses.

However, OpenAI’s launch materials described the model as trained on images and text reflecting a wide range of styles. They did not establish that particular Studio Ghibli films, frames, or production materials were included in the training set. It would therefore be inaccurate to state as fact that OpenAI trained on Ghibli films.

The U.S. Copyright Office has treated this as a separate policy and legal issue. Its January 2025 report, Part 2, addressed the copyrightability of AI-generated outputs. The initiative identified Part 3 as addressing the legal implications of training models on copyrighted works, including licensing and potential liability. The structure of that work illustrates why training could not be reduced to a settled rule that applies automatically in either direction.

Rank #3

Sale

The Art of Howl's Moving Castle

• New
• Mint Condition
• Dispatch same day for order received before 12 noon
• Guaranteed packaging
• No quibbles returns

$16.59

Buy on Amazon

2. Did an individual output copy protected expression?

Even if a model had learned from copyrighted material, the appearance of a particular output would still require its own analysis. Copyright generally protects expression rather than an abstract idea, technique, or broad aesthetic category. TechCrunch quoted attorney Evan Brown explaining that style is not expressly protected by copyright and that an image merely resembling the look associated with Studio Ghibli would not automatically be unlawful on the reported facts.

That does not make every output safe. “Style” can be a loose word for elements that are more specific and recognizable. An image that reproduces a protected character, a distinctive setting, a recognizable composition, or a near-identical film frame raises a more substantial copying or derivative-work question than an original scene that shares only a gentle color palette, hand-drawn atmosphere, or fantastical subject matter.

This is a spectrum, not a legal test with automatic results. The outcome could depend on the exact prompt, the output, the source material, the model’s behavior, the degree of similarity, the parties involved, and the court hearing the dispute.

Where fair use fits

OpenAI has maintained that training its models can qualify as fair use. Fair use is a potential defense, not a declaration that a particular training practice or output is lawful.

A court applying the U.S. fair-use doctrine would generally examine the purpose and character of the use, the nature of the copyrighted works, the amount and substantiality used, and the effect on actual or potential markets. Applying those factors to large-scale model training is disputed. The analysis may involve the copying needed to develop the model, the functions the model performs, whether the use is transformative, and whether the system competes with or substitutes for licensed creative work.

Rank #4

Sale

The Art of My Neighbor Totoro: A Film by Hayao Miyazaki

• New
• Mint Condition
• Dispatch same day for order received before 12 noon
• Guaranteed packaging
• No quibbles returns

$15.53

Buy on Amazon

The same defense does not automatically answer an output dispute. Training and generating an image are related events, but they involve different acts and different evidence. Nor does the fact that an output is newly generated prove that the training process was lawful. Conversely, a model’s ability to produce a Ghibli-like image does not prove that a particular film was copied into the model or that a specific output infringes.

The fake cease-and-desist document

During the controversy, a purported Studio Ghibli cease-and-desist letter circulated online. Futurism reported that Studio Ghibli told NHK the document was fake.

This correction is important because the document was sometimes treated as evidence that Ghibli had already made a formal legal demand. It was not reliable support for that claim. The fact that Ghibli declined to comment to AP during the initial reporting should also be described accurately: it documented the studio’s lack of public comment at that time, not an admission that it approved the images or had no legal objection.

What happened after the original image trend?

Later reporting in October 2025 described a request from the Content Overseas Distribution Association, or CODA, a Japanese trade organization representing companies including Studio Ghibli. CODA reportedly asked OpenAI to stop using members’ copyrighted works to train its AI systems, particularly in connection with Sora 2. The request reportedly warned that unauthorized reproduction during machine learning could create legal exposure.

Best Value

Sale

The Art of Princess Mononoke

$18.59

Buy on Amazon

That development reinforces the importance of the training-data question, but it should not be conflated with a confirmed Studio Ghibli lawsuit arising from the March 2025 ChatGPT image trend. A trade-organization request is not the same as a court complaint, judicial ruling, settlement, or finding that OpenAI infringed Ghibli’s rights.

Why the lawyer’s wording matters

The careful reading of the headline is therefore:

1. OpenAI released a native image-generation feature in GPT-4o.
2. Users quickly used it to create images associated with Studio Ghibli’s visual identity.
3. A lawyer identified possible Lanham Act, trademark, false-endorsement, unfair-competition, and copyright theories.
4. The strength of those theories depended on facts that were not publicly established, including OpenAI’s training data, any license, its marketing language, and the similarity of particular outputs.
5. The original reporting did not establish that Studio Ghibli had sued OpenAI or that OpenAI had been found liable.

The controversy exposed a genuine legal fault line: a company may face different risks when it markets a tool using another brand’s goodwill, when its model was allegedly trained on copyrighted works, and when a user generates an image containing recognizable protected expression. Treating all three as one question produces a much stronger claim than the evidence supports.

Sources and scope

• Futurism, March 28, 2025 report quoting Rob Rosenberg on possible legal action and the purported cease-and-desist document.
• OpenAI’s March 25, 2025 announcement of GPT-4o image generation and its description of image transformation, style handling, and training material.
• TechCrunch reporting on users creating Studio Ghibli-style images and the distinction between style and protected expression.
• Associated Press reporting on the unanswered licensing and training-data questions, as well as recognizable characters, scenes, and other expressive elements.
• U.S. Copyright Office materials from its Copyright and Artificial Intelligence initiative, including the January 2025 Part 2 report and the identified Part 3 analysis of training on copyrighted works.
• October 2025 reporting on CODA’s request concerning copyrighted works and AI training.

This is a legal-news explainer, not legal advice. The Lanham Act analysis concerns U.S. law; rights and remedies may differ in Japan and other jurisdictions.