Studio Ghibli, CODA, and Sora 2: How to Run a Coalition Demand Letter Against an AI Giant
When Studio Ghibli, Square Enix, Bandai Namco, Toei and other Japanese rights holders tell OpenAI to stop using their content, they’re not just complaining about “AI stealing art.”
They’re running a coordinated, trade-association demand campaign that:
- targets a specific product (Sora 2),
- attacks both its training data and outputs, and
- leverages a legal system (Japan) that is much less friendly to “train now, ask forgiveness later.”
For anyone who drafts demand letters, licenses IP, or negotiates AI/data clauses for tech clients, the CODA–Sora 2 situation is basically a live demo of how to structure a coalition demand letter against a platform.
Who Is CODA and Why Is Sora 2 in Their Crosshairs?
The Content Overseas Distribution Association (CODA) is a Japanese anti-piracy and content-protection trade group representing 30+ major IP owners, including Studio Ghibli, Bandai Namco, Square Enix, Toei Animation, Aniplex, and major music labels. (GamesRadar+)
OpenAI’s Sora 2 is a video-generation model that went viral for producing short clips that look suspiciously like anime, JRPG cinematics, and other Japanese pop-culture styles. Japanese media and watchdogs quickly flagged Sora 2 clips that, to the naked eye, appear to replicate well-known characters and visual language from anime and games. (Windows Central)
CODA’s official letter, dated October 27, 2025, makes three core assertions: (coda-cj.jp)
- A “large portion” of Sora 2’s outputs closely resembles Japanese content and images.
- CODA believes this results from using members’ works as training data without authorization.
- Under Japanese law, prior permission is required; OpenAI’s “opt-out” approach is not enough.
They are not (yet) filing suit. They are telling OpenAI:
Stop using our members’ works in Sora 2 training, and answer our copyright questions honestly.
In other words, this is demand-letter phase, not litigation phase—but it’s structured with litigation and regulation in mind.
What CODA’s Letter Actually Demands
CODA’s English-language statement and media summaries are remarkably concrete for a public-facing letter. (coda-cj.jp)
Here is the core posture in one view:
| 🏢 Sender side (CODA & members) | 🧠 What they’re saying | 🎯 Practical demand |
|---|---|---|
| Studio Ghibli, Square Enix, Bandai Namco, Toei, music labels and others, acting via CODA 🎌 | Sora 2 outputs “closely resemble” protected Japanese works, and this likely reflects unauthorized use of those works in training. | Stop using member content in training datasets unless you get prior permission. |
| Japanese IP as cultural assets 🎎 | Popular anime, manga and game characters are “irreplaceable treasures” that Japan presents to the world; treating them as free training fodder is an attack on cultural and economic value. (Reddit) | Recognize that this is not generic “style”; it’s specific IP tied to national cultural capital. |
| Legal theory under Japanese law 📜 | Training that reproduces or “similarly generates” copyrighted works may itself be copyright infringement; opt-out systems conflict with the requirement of prior authorization. (coda-cj.jp) | Abandon “opt-out” and do not rely on it as a shield; secure affirmative licenses. |
| Transparency demand 🔍 | OpenAI must “respond sincerely” to copyright inquiries and explain what data was used and how outputs relate to training sets. (Tech Times) | Provide detailed answers about training corpora, filtering, and any steps taken to avoid Japanese IP in outputs. |
From a demand-letter drafter’s viewpoint, notice the structure:
- Specific product: Sora 2, not “all OpenAI models ever.”
- Specific members: 30+ named or easily identifiable rights holders, not a vague “artists of the world” coalition.
- Specific legal theory: Japanese copyright law requires prior permission; opt-out is invalid.
- Specific asks: cease training on member content, explain what has already been done, and engage on copyright complaints.
This is not just moral outrage; it’s a lawsuit-ready fact pattern on letterhead.
The Legal Backdrop: Japan’s Copyright Exceptions vs Prior Permission
One reason CODA’s letter is so pointed is that Japan’s copyright law is genuinely different from the U.S. in how it treats AI training.
Japan has a “non-enjoyment purpose” exception (often discussed under Article 30-4) that allows certain data-mining and AI training uses of copyrighted works without consent, but with important qualifiers: (Privacy World)
- The exception is aimed at non-enjoyment purposes (data analysis, R&D), not reproducing the expressive content in a consumer-facing way.
- It becomes much shakier when training is directed toward reproducing creative expressions, like recognizable characters or scenes, or when there is material impact on the market for the works.
CODA’s argument is essentially:
Sora 2 is not just “analyzing” our works. It is replicating them (or near-copies) in outputs. That means the machine-learning replication and the outputs are both infringing uses that require prior permission.
That meshes with commentary from Japanese practitioners:
- They note that the Article 30-4 exception does not bless training whose purpose is to recreate expressive elements, especially in a way that competes with the original works or merch. (Global Practice Guides)
So CODA is not just complaining about style mimicry; it’s building a case that Sora 2’s training and outputs fall outside Japan’s AI exception and back into ordinary infringement territory.
Coalition Demand Letters as a Strategy, Not a Press Stunt
What makes the CODA–Sora 2 situation especially interesting is the coalition structure.
Instead of each studio firing its own lawyer letter, they are:
- acting through a trade association,
- coordinating messaging with government officials, and
- running the campaign in the open.
Japanese officials have publicly echoed similar concerns, calling anime and game characters “irreplaceable treasures” and urging OpenAI to refrain from infringing uses. (Reddit)
From a strategy standpoint, a coalition demand letter does at least three things a solo letter cannot:
| 🤝 Feature | ⚖️ Legal effect | 💥 Practical pressure |
|---|---|---|
| Aggregated rights holders | Makes it plausible that training sets contain hundreds or thousands of implicated works, not a single title. | Multiplies potential damages and makes “this is a rounding error” harder to argue. |
| Unified legal theory | Shows a consistent reading of local law (prior permission, non-enjoyment limits) that courts and regulators can pick up. | Reduces the risk that defendants can play studios off against each other. |
| Cultural framing | Positions Japanese works as national cultural assets, not just random copyrights. | Raises reputational stakes for OpenAI in a key market; makes legislators more likely to get involved. |
And crucially: CODA’s letter is public. It’s designed to be read by judges, regulators, and other tech companies, not just by OpenAI’s legal team.
Anatomy of the CODA–Sora 2 Demand
If you treat CODA’s statement as a template, the bones of a strong coalition demand letter look like this:
| 🧩 Element | 📝 How CODA does it | 📌 Takeaway for your own coalition letters |
|---|---|---|
| Clear identification of sender | CODA speaks explicitly on behalf of 30+ named members (Ghibli, Square Enix, Bandai, Toei, music labels, etc.). (GamesRadar+) | Don’t hide behind abstractions; list categories of rights holders and key flagship names. |
| Concrete factual trigger | Sora 2 outputs “closely resembling” Japanese works after its launch, including videos mimicking famous anime/game franchises. (coda-cj.jp) | Tie your complaints to a particular model, release, or feature, with observable examples. |
| Jurisdiction-specific legal theory | CODA cites Japanese copyright law’s prior-permission requirement and argues that both training replication and similar outputs “may constitute copyright infringement.” (coda-cj.jp) | Anchor arguments in the strongest jurisdiction you have, even if the company is U.S.-based. |
| Attack on the opt-out model | CODA labels OpenAI’s opt-out policy legally incompatible with Japanese law; “there is no system allowing one to avoid liability” merely by opting out later. (GamesRadar+) | Don’t accept the platform’s preferred “global” legal framing; show why their UX/policy is invalid where your members live. |
| Specific demands and a tone of “we’re willing to talk” | The letter demands an immediate stop to using member content without authorization and a sincere, transparent response to copyright inquiries. (Tech Times) | Keep asks concrete: stop certain conduct, disclose information, and open a channel to negotiate licensing or technical fixes. |
If you’re advising a guild, trade group, or collection of smaller creators who want to confront a platform, this is a pattern you can adapt.
Lessons for IP Owners Outside Japan
Even if you’re representing U.S. or EU rights holders, CODA’s move gives you a playbook.
A few practical points that translate well:
- Organize first, write later. A coalition letter from a trade body or joint signatories gets more attention than 50 solo letters scattered over six months.
- Pick your best forum. CODA leans into Japan’s prior-permission framework and AI-exception limits; you can do something similar with EU text-and-data mining rules or state-level publicity/privacy law. (arXiv)
- Treat the letter as both legal document and political messaging. CODA talks about “treasures” and cultural exports for a reason; that language is for ministries and voters as much as for OpenAI.
- Think beyond money. CODA is not opening with “here is our price per frame.” They’re opening with: stop, explain, and then we can talk. That sequencing builds a record for injunctions or regulatory action if negotiations fail.
You can plug the same logic into a Terms.Law style demand-letter generator:
- identify the coalition,
- pick the jurisdiction and legal theory,
- define a target AI product or model,
- generate a tailored set of demands (cease training, filter outputs, disclose training data sources, discuss licensing).
Lessons for AI and SaaS Companies on the Receiving End
For AI builders and data-hungry SaaS companies, CODA’s letter is the nightmare version of an “angry creator email”:
- It’s organized.
- It’s jurisdiction-specific.
- It has government and press amplification.
If you’re advising one of these companies, the practical guidance looks something like this:
- Treat coalition demand letters as board-level events. This isn’t a routine DMCA notice; it implicates your core training pipeline and market access in an entire country.
- Map your exposure by jurisdiction. If your training stack relies heavily on Japanese content, you need a plan for:
- pausing or limiting training on that corpus,
- tightening output filters, and
- preparing a credible explanation of how your data-governance and opt-out systems work.
- Assume the letter will show up in a regulator’s file. CODA’s demands and the Japanese government’s public comments are already being cited in AI governance analysis; you should respond as if an agency and a judge will eventually read the full correspondence. (IAPP)
From a contract perspective, it is also a reminder to:
- avoid vague, global “we may train on publicly available content” provisions in your own ToS when you’re operating in jurisdictions with stricter rules;
- differentiate training rights by territory, and consider explicit carve-outs or affirmative licenses for places like Japan where prior permission is the expectation.
Where Coalition Demand Letters Fit in the AI Dispute Lifecycle
CODA’s engagement with OpenAI sits at an interesting point on the escalation ladder:
| 📶 Stage | What’s happening | CODA–Sora 2 example |
|---|---|---|
| 1. Quiet outreach | Direct contact, private negotiation, NDAs, etc. | We don’t see this; if it happened, it didn’t solve the problem. |
| 2. Coalition demand letter 📮 | Formal letter from trade group or joint signatories; often public or leak-ready. | CODA’s October 27 letter and public statement about Sora 2’s training and outputs. (coda-cj.jp) |
| 3. Regulatory pressure 🏛️ | Ministries and agencies issue guidance, warnings or open investigations. | Japanese officials publicly warn OpenAI about infringing anime/game characters and stress their cultural value. (Reddit) |
| 4. Litigation ⚔️ | Lawsuit (often test case or class action) in a favorable forum. | Not yet in Sora 2’s case; CODA’s letter is clearly designed so a complaint could follow. |
| 5. Structured licensing / settlement 🤝 | Portfolio deals, consent decrees, or structured licensing schemes (per-work, per-minute, per-stream). | Where this likely heads if OpenAI wants to keep a meaningful Japanese presence without protracted fights. |
If you do corporate or tech work, you can position yourself at stages 2–5:
- Stage 2: drafting coalition demand letters that don’t just vent, but set up a litigation and negotiation runway.
- Stage 3: helping clients respond to or leverage regulatory guidance in their contracts.
- Stage 4: preparing complaints anchored in the facts and legal theories telegraphed in the letter.
- Stage 5: negotiating cross-portfolio licensing deals or structured settlement agreements that define when and how content can be used for AI.
Takeaways for Your Own Practice
The Studio Ghibli / CODA vs OpenAI Sora 2 story is not just a headline about anime and AI. It’s a working example of:
- how to aggregate rights holders into a single, credible voice;
- how to localize legal theory to the most favorable jurisdiction; and
- how to use a public demand letter as both a legal instrument and a political document.
If you’re advising creators, publishers, or trade groups, this is a model for building a coalition that can actually force a negotiation with a platform.
If you’re advising AI or SaaS companies, it’s a reminder that “we have an opt-out” is not a universal defense—and that ignoring or downplaying a coalition demand letter is a fast way to become the test case everyone else cites.