Japan has reignited the global debate over intellectual property and artificial intelligence. The Japanese government has formally asked OpenAI to stop using manga and anime works without consent to train its video model Sora 2, a tool capable of generating hyperrealistic animated clips in seconds.
According to The Japan Times, the country is demanding that the company adopt an opt-in model—in which creators explicitly authorize the use of their works—instead of the opt-out system OpenAI currently employs.
Beyond this specific case, the episode reopens an urgent question: To what extent can AI “learn” from culture without appropriating it?
Learning isn’t copying… or is it?
Tech companies argue that their models don’t store copies of works but rather extract statistical patterns. Artists, on the other hand, claim those patterns are part of their style and therefore their authorship.
At its core, the conflict is not legal but philosophical: Does creation belong to the individual or to the community that inspires them?
Artificial intelligence has challenged the copyright model born in the 18th century to protect printed books—not neural databases. Today, training an AI is more like educating it than plagiarizing it, yet the law still treats it as copying.
As analyzed in another Emprender y Más article on how ethics drives technological innovation, we keep applying rules of the past to problems of the future.
The Anthropic case: when learning costs millions
While regulators debate, some giants have already chosen to pay.
According to Axios, Anthropic—the creator of the Claude model—agreed to compensate U.S. writers and media outlets with $1.5 billion for texts used to train its systems without permission.
The deal not only sets a legal precedent—it inaugurates a market where creativity becomes training currency.
Every paragraph, every song, and every image used to “teach” an AI now has a monetary value.
But the most relevant question isn’t how much Big Tech will pay, but how small entrepreneurs will compete if every dataset becomes unaffordable due to licensing costs.
A broken model: property vs. knowledge
Intellectual property was created to encourage individual creation, not to restrict collective learning. Today, its rigid application risks stifling innovation: if all knowledge is owned, no one can learn.
That’s why many legal and tech experts propose a new legal category for “machine learning use”—a sort of intermediate license between educational and commercial use.
This framework would allow AIs to learn freely from protected works as long as the use isn’t directly profit-driven, reserving rights for commercial applications.
The challenge, of course, is defining what counts as “learning” and what counts as “exploitation.”
From YouTube to AI: the missing revenue-sharing model
Digital platforms have already solved similar dilemmas.
On YouTube, if a video uses protected music, the platform automatically identifies the content and shares part of the ad revenue with the rights holder. On Spotify, each stream generates micro-payments distributed between labels and artists.
AI could do the same: identify works used in training (through fingerprinting or blockchain), share a symbolic fraction of the value generated when results are used commercially, and free up research or educational uses, as academic fair use does.
It’s not about banning or pirating—it’s about monetizing without blocking innovation.
YouTube didn’t destroy the music industry; it transformed it. Spotify didn’t erase authorship; it scaled it. AI could follow the same path if it turns mass usage into shared value.
A premium model for sustainable creativity
The solution may lie not in fines, but in the model itself. If knowledge is the new fuel, AI needs a creative economy that works like an ecosystem, not a mine.
One possibility would be a tiered access model, similar to streaming platforms.
Free AI use would allow content generation with public-domain or open-licensed material. But to access protected styles, images, or voices, users would need a premium account, whose fees would finance royalties for original authors.
The system would be simple: each time an AI uses datasets with registered works, a percentage of the user’s payment goes to a rights fund. Creators receive compensation proportional to the real use of their works. Data traceability (through blockchain or digital watermarking) would ensure fair and transparent distribution.
This approach would turn machine learning into a constant source of income for humans, without halting innovation. Instead of punishing use, it would reward talent.
More than laws, we need culture
The heart of the problem isn’t just copyright—it’s the culture of control. For centuries, creativity was understood as something to be possessed, but AI confronts us with an uncomfortable truth: creation has always been a collective act.
Algorithms merely accelerate what humans have always done: take inspiration, combine, and reinterpret what others have created. The challenge is how to recognize and compensate those influences without slowing the evolution of knowledge.
As noted in this analysis on innovation and cooperation, the most sustainable models are those that turn competition into collaboration.
Toward a new creative contract
Japan protects its cultural heritage; Anthropic pays to learn; Europe legislates; and entrepreneurs watch in uncertainty. Perhaps it’s time to imagine a new contract between creators, companies, and machines—one where knowledge is shared, but value is distributed.
The future of intellectual property will not be about exclusion, but about traceability and participation. AI doesn’t need to be the enemy of artists; it can become their most profitable partner.
Because if creativity is the fuel of algorithms, progress will depend on paying for the gasoline, too.
