Authorship’s Future: AIPPI’s 2025 Resolution and AI-Era Copyright

AIPPI Resolutions and Their Significance

The International Association for the Protection of Intellectual Property (AIPPI) plays a unique role in shaping international IP law. Although its resolutions are not legally binding, they reflect a carefully negotiated consensus of IP experts, judges, and practitioners across jurisdictions. Over time, AIPPI’s resolutions have influenced both international treaties and national legislative reforms, often serving as the intellectual groundwork for harmonisation.

The 2025 Resolution on AI & Copyright continues this legacy. It sets forth guiding principles for addressing the tension between the need for large-scale data to train AI systems and the legitimate interests of creators whose works form part of that data.

Key Provisions of the Resolution

Training AI Systems on Copyrighted Works Generally Requires Prior Authorisation

 AIPPI resolved that, as a general rule, the use of a copyrighted work to train an AI System should require the prior authorisation from the copyright holder, unless such use is covered by one or more specific exceptions.

Recognised Exceptions to the Authorisation Requirement

1. Existing copyright exceptions may apply. The use of copyrighted works to train AI systems should be subject to the same exceptions that apply to other uses of copyrighted material—whether or not for commercial purposes—if the statutory conditions are met.
2. A new public-interest exception for non-profit research and education. training that is not-for-profit and conducted solely in the public interest, such as for scientific research or education, may be permitted without authorisation. This exception “does not extend to commercially exploiting the trained AI System or the training dataset.”
3.  Commercial training without consent must allow opt-out and compensation. If a jurisdiction permits commercial AI training without prior authorisation, right-holders must have the right to opt out, and where they do not exercise that right, financial compensation must be provided.
4. All exceptions must comply with the Berne Convention’s three-step test. Each exception must be a special case, must not conflict with the normal exploitation of the work, and must not unreasonably prejudice the author’s legitimate interests.

Together, these provisions aim to strike a careful balance between enabling AI development and safeguarding authors’ economic and moral interests.

Transparency Obligations Aim to Lift the Veil on AI Training Datasets

A significant innovation of the resolution is the introduction of transparency obligations for AI providers and trainers.

The resolution provides that they must provide adequate information regarding the copyrighted works used in the training of the AI System to enable copyright holders to identify the use of their works and exercise or enforce their rights.

In addition, they must identify any copyrighted materials input by a user to the AI System and used by the AI System for training.

These measures address growing concerns about the opacity of AI training datasets and aim to ensure that creators can monitor and enforce their rights effectively.

AI Outputs Are Not Automatically Infringing Simply Because They Resemble Human Works

The resolution clarifies that the rules of an applicable jurisdiction to determine copyright infringement should also apply to output from a trained AI System.

It then elaborates on several specific aspects:

• Moral rights: Authors should have the right to object to an output of an AI System that constitutes a mutilation, distortion or other derogatory action that would prejudice their honour or reputation.
• Style imitation: An AI output should not constitute copyright infringement for the sole reason that it is in the same style as a copyrighted work used to train an AI System.
• Training infringement: An output should not constitute infringement… for the sole reason that training the AI System has infringed copyright.
• Scope of exception or authorisation: When an AI system has been trained under an exception or licence, whether its outputs infringe depends on the scope of that exception or licence. If it covers only training, the outputs may still infringe; if it extends to outputs, those outputs are exempt.

This nuanced approach separates the legality of training from that of outputs and introduces proportionality into infringement assessment.

AI Systems Themselves May Be Deemed Infringing in Cases of Unlawful or Deliberately Infringing Training

In a groundbreaking statement, the resolution declares that an AI System itself should be considered an infringing article in two circumstances:

1. where more than a de minimis amount of the training was conducted with copyrighted materials used unlawfully; or
2. where the AI system has been developed specifically to create infringing outputs.

This provision introduces the possibility of treating an AI model as a product embodying infringement, thereby allowing direct enforcement measures against it.

Liability Extends Beyond Developers to Exploiters and Users

The resolution provides that, depending on the facts of a case, one or more of the following parties may be held liable for infringing outputs:

1. The provider of the AI system, defined as the entity that develops and/or places it on the market;
2. Any person or entity that commercially exploits the AI system; and
3. Any person or entity that uses the AI system with the aim of creating infringing outputs, for example, through detailed and deliberate prompting.

This formulation accommodates the distributed nature of AI ecosystems, recognising that responsibility may lie with multiple actors.

Effective Remedies Must Compensate and Deter

The resolution adopts a comprehensive approach to remedies. If copyright is infringed through training, outputs, or because the AI system itself is infringing, right-holders are entitled to damages, injunctive relief, recall from commercial channels, and destruction of infringing materials or systems. Damages and account of profits should compensate the copyright holder for damages caused by the infringement and the bypassing of consent.

All remedies must be proportionate, effective, and deterrent, ensuring meaningful protection for right-holders.

Conclusion

AIPPI’s 2025 Resolution on AI & Copyright stands as a milestone in the ongoing dialogue between technology and creativity. Its detailed framework—rooted in the language of existing copyright law yet forward-looking in its treatment of AI—offers a pragmatic blueprint for legislators worldwide.

The Yokohama Resolution is not merely an academic exercise; it is a call for coherence, fairness, and transparency in the age of generative AI—and a major step toward reconciling human authorship with machine learning in the modern copyright landscape.

***

Ran Vogel is a partner in the firm’s IP Group and is available to address questions on Copyright and AI, as well as other intellectual property matters.