Tune worth borrowing?

As artificial intelligence (AI) increasingly weaves itself into everyday life, it raises novel and thorny legal questions. One of them is whether AI may freely use copyrighted works in the course of learning and generating responses.

In GEMA v OpenAI (Case 42 O 14139/24, 11 November 2025), the Munich I Regional Court ruled that OpenAI infringed German copyright law when its GPT-4 and GPT-4o models memorized and reproduced copyrighted song lyrics.

GEMA, the German society representing songwriters, composers and music publishers, sued OpenAI over the lyrics of nine popular songs, including Kristina Bach’s “Atemlos” and Herbert Grönemeyer’s “Männer.”

When prompted, ChatGPT reproduced the lyrics almost entirely, with only minor deviations. GEMA consequently sought injunctions and other relief against OpenAI.

OpenAI maintained that its models merely learn statistical patterns rather than store copyrighted texts. It argued that the outputs resulted from prompts deliberately designed to elicit the lyrics and that any copying fell within the text and data mining (TDM) exceptions under German law.

The Court was not persuaded.

It distinguished three phases of AI development: building the training dataset, training the model and generating outputs in response to prompts.

According to the Court, the models had memorized the lyrics during training. The prompts were too simple, and the reproduced lyrics too extensive, to be dismissed as coincidence.

The Court reasoned that storing copyrighted material in numerical form does not necessarily remove it from copyright protection. After all, an MP3 file remains a reproduction of a song despite its digital format.

Neither could OpenAI successfully invoke the TDM exception. The Court distinguished permissible data analysis from memorization that results in the reproduction of protected works. It likewise rejected the scientific research exception, considering OpenAI’s commercial character.

Significantly, however, the Court did not declare that training AI models using copyrighted materials is inherently unlawful. Its ruling focused on the memorization and reproduction of identifiable protected works.

The Court also held OpenAI directly liable for the infringing outputs, rejecting the argument that responsibility rested solely with users who entered the prompts.

Consequently, OpenAI was ordered to cease the infringing reproductions, account for relevant revenues and compensate GEMA for damages, subject to the judgment’s terms.

But the decision should not be mistaken for a declaration of war against artificial intelligence.

AI is a transformative technology whose enormous potential for education, research, business and even the administration of justice cannot be ignored. Copyright protection is important, but so is technological innovation. The challenge is to strike a reasonable balance between the legitimate rights of creators and the equally legitimate advancement of technology.

In the Philippines, where no AI-specific copyright framework or express TDM exception presently governs the issue, the question would have to be examined under our Intellectual Property Code, including its provisions on fair use.

The German decision, of course, is not binding on Philippine courts. Neither should its reasoning automatically dictate how our own copyright laws must be interpreted.

Will the Filipino Society of Composers, Authors and Publishers eventually take the same route as GEMA? More importantly, will Philippine courts adopt the same interpretation or chart a different course that affords greater latitude to AI innovation?

Your guess is as good as mine. For now, one thing is certain: the law must protect creativity without stifling the very technology that may inspire the next generation of creators. After all, the law should not merely preserve yesterday’s music. It should also make room for tomorrow’s melodies

For more of Dean Nilo Divina’s legal tidbits, please visit www.divinalaw.com. For comments and questions, please send an email to cad@divinalaw.com