Source materials can be found here.

AI and Copyright at WIPO: Key Takeaways for Scholarly Publishers

by Claude

Live from Geneva: The World Intellectual Property Organization's recent Standing Committee on Copyright and Related Rights (SCCR) tackled the intersection of artificial intelligence and copyright law – with major implications for academic publishing.

The marble halls of WIPO's Geneva headquarters buzzed with tension this week as government representatives, tech giants, creators, and legal experts gathered to address one of the most pressing intellectual property questions of our time: How should copyright law respond to generative AI?

For scholarly publishers, the discussions revealed both challenges and opportunities as stakeholders from across the globe shared their approaches to AI training data and AI-generated outputs. Here's what you need to know.

The Great Training Data Debate

Perhaps the most consequential issue for publishers is whether AI companies need permission to use copyrighted materials to train their models. The divide was striking.

Google's Cedric Manara argued that requiring permission would have stifled innovation: "If we would have had to obtain authorization from every single content owner, then Google search engine would not exist today." He compared AI training to search engine indexing, suggesting similar exceptions should apply.

Many jurisdictions are considering different approaches:

• The European Union has implemented text and data mining exceptions with an opt-out mechanism allowing rights holders to reserve their works.

• Japan allows text and data mining without permission under Article 30/4 of their Copyright Act but specifically prohibits collecting data to generate similar works for "enjoyment purposes."

• Brazil is moving in a different direction with a draft decree requiring transparency and obligatory remuneration for rights holders whose content trains AI.

For scholarly publishers, Emmanuel Duchar of the European Commission offered perhaps the most relevant insight, emphasizing that successful policies must "ensure the development of licensing markets between the creative industries and AI providers."

Is AI-Generated Content Copyrightable?

The second major question addressed whether AI-generated content deserves copyright protection – a critical consideration as we see increasing submissions of AI-assisted academic papers.

Most jurisdictions appear aligned on human creativity as a prerequisite for copyright protection:

• Korea's approach, as described by Director Jung, makes "human creative contribution the decisive factor in determining whether an AI-assisted work can be protected by copyright."

• Georgia's IP Center has refused copyright registration to AI-generated works, applying a "first step test" to assess human intellectual effort.

• The United States position was referenced through the "Thaler v. Perlmutter" case where courts ruled only human work can be copyrightable.

However, Professor Qian from East China University highlighted a contrary approach in China, where courts have granted copyright to AI-generated images when humans design the prompts. She disagreed with these rulings, arguing, "The generative AI such as deep seek can produce content based on its own algorithm and data training. The user's input cannot really decide the resulting content."

Detection and Distinction

A fascinating development for publishers was Deezer's announcement of technology to identify AI-generated music. Their system revealed that "10-15% of daily music content delivered to platform is 100% AI-generated," according to Tibo Boudin, who also noted that Deezer has "blocked AI content from being recommended to users."

Such detection capabilities could prove valuable for scholarly publishers grappling with unattributed AI-generated submissions. Boudin's revelation that despite comprising 10-15% of uploads, this content represents only 0.1% of actual listening suggests audiences still prefer human-created work – potentially reassuring news for scholarly publishing.

What Creators Want

Creative stakeholders repeatedly called for fair compensation and recognition. Aaron Chaturvedi of the Songwriters Association of Canada emphasized that for creators, copyright is "how we get paid and provide for our families," while warning of "declining local economies as global tech giants secure financial dominance."

Several stakeholders requested WIPO lead an expert study on legislative options to protect human authors, with Aaron Chaturvedi memorably calling for AI to "move fast and fix things, not move fast and break them."

A Middle Path Emerging?

Some innovative approaches suggest potential paths forward. Anastasia Stasenko from Players AI described creating a "common corpus" of 2 trillion words with proper licensing attribution for training AI models, demonstrating that responsible AI development needn't rely on unauthorized use of copyrighted materials.

Meanwhile, Abhishek Singh from India's Ministry of Electronics and IT described efforts to develop "a model framework that can serve as template for Global South countries" – balancing IP protection while allowing innovation.

Implications for Scholarly Publishers

For scholarly publishing, several key considerations emerge:

1. Training data licensing will likely become more structured, potentially creating new revenue streams for academic content

2. Human authorship requirements will remain central to copyright protection, supporting publishers' need for attribution and accountability

3. Detection technologies for AI-generated content will continue to improve, assisting editorial screening processes

4. Legal fragmentation across jurisdictions creates compliance challenges for international publishers

5. New licensing frameworks between publishers and AI developers represent a potential growth area

WIPO Deputy Director General Sylvie Forbin concluded the session by acknowledging the risks of fragmentation and revealing that WIPO is developing a new platform dedicated to addressing these issues, to be announced at the upcoming WIPO Conversation on IP and AI.

As one Brazilian delegate noted, this may indeed be "the mother of all copyright debates." For scholarly publishers navigating an increasingly AI-influenced landscape, staying engaged with these developments will be essential to protecting intellectual assets while embracing innovation.

AI Takes Center Stage at WIPO SCCR 46: Insights for Scholarly Publishers

by Gemini Advanced 2.5 Pro

Reporting directly from the 46th Session of the WIPO Standing Committee on Copyright and Related Rights (SCCR) in Geneva, it's clear that Artificial Intelligence (AI), particularly generative AI, is no longer a peripheral topic but a central concern demanding urgent attention within the international copyright community. While the committee tackled its traditional agenda items like broadcasting rights and limitations and exceptions (L&Es), the dedicated information session on AI and numerous interventions throughout the week highlighted the profound questions AI poses for creators, rights holders, and the future of the copyright framework itself.

For scholarly publishers navigating this rapidly evolving landscape, here are the key takeaways and clustered themes from the SCCR discussions:

The Core Conundrum: Training Data & Copyright

The dominant theme was the relationship between generative AI model training and copyright law.

• Input vs. Output: A critical distinction was drawn between using copyrighted works to train AI models (input) and the material the AI subsequently generates(output). Professor Peter Meze (Hungary) set the stage, noting AI training is deeply intertwined with technology, law, and policy, requiring a delicate balance between protecting creative industries and fostering innovation.

• Legal Basis for Training: Panelists debated the legality of using copyrighted works for training.

  • Exceptions & Limitations: Several jurisdictions, including the EU and Japan, discussed their Text and Data Mining (TDM) exceptions [cite: 1950-1953, 1980]. Emmanuel Duchar (European Commission) explained the EU's approach, which includes an opt-out mechanism for rights holders, though its practicality and standardization remain challenges. Hirohiko Nakahara (Japan Copyright Office) outlined Japan's TDM exception, noting it doesn't apply if the use is for the 'enjoyment purpose' of the work or unduly harms rights holder interests.

  • Fair Use: Mary Critharis (USPTO) highlighted that in the U.S., AI companies often invoke the fair use doctrine as a defense against infringement claims related to training data, though no court has made a final ruling on this yet [cite: 2052-2055, 2057-2059]. Cedric Manara (Google) compared AI training to search engine indexing, arguing that requiring permission for all web content would stifle innovation, suggesting training falls under fair use [cite: 1885-1889, 2201].

  • Rights Holder Concerns: Representatives from creative sectors (like the European Writers Council, FILAE, ADAGP, IAWG, FIA, IFPI) expressed strong skepticism about applying exceptions like TDM or fair use to AI training, viewing it as mass infringement without consent or compensation. They highlighted the difficulty of exercising opt-outs effectively due to lack of transparency and standardization [cite: 2228, 2348, 2349, 2386-2390].

• Transparency: A recurring demand from rights holders and some Member States (e.g., Brazil, India, EU via the AI Act) was for greater transparency regarding the data used to train AI models. Anastasia Stasenko (Players AI) noted that major models have trained on pirated content and web archives containing copyrighted material, making compliance difficult without provenance [cite: 2507-2510, 2514, 2515]. Her company focuses on building models using demonstrably licensed or public domain data.

• Licensing & Remuneration: Many participants saw licensing as a potential path forward.

  • Google and Shutterstock noted an emerging licensing market and collaborations between AI developers and content owners.

  • Brazil's draft bill mandates remuneration, potentially facilitated by Collective Management Organizations (CMOs). Marcos Alvestizuza (Brazil) stressed this is crucial to prevent the copyright system's collapse.

  • The practicality of different remuneration models (e.g., per use, collective) and handling non-CMO members were questioned.

  • Concerns were raised about the bargaining power imbalance between tech giants and creators/publishers.

AI-Generated Outputs: Authorship & Infringement

The status of content produced by AI was another major focus.

• Copyrightability: There's a general consensus that works generated solely by AI (without human intervention) are not protectable by copyright under current frameworks, which require human authorship. Professor Wang Qian (China) argued such outputs should enter the public domain.

• AI-Assisted Works: For works where AI assists a human creator, the question of copyright protection hinges on the level of human creative input and control [cite: 2248-2252, 2558, 2559]. Korea shared its experience registering the first AI-assisted work based on substantial human refinement, while Georgia detailed refusing deposition for poems lacking sufficient human contribution. Professor Qian argued user prompts are merely ideas and don't grant authorship over the output if the AI's process is a "black box" [cite: 2655-2661, 2670, 2671, 2924-2927, 2932-2934]. Graciela Melo Sarmiento (Colombia) emphasized originality requires freedom and choice, which is arguably absent in the AI generation process.

• Infringement & Liability: Determining if an AI output infringes an existing work involves traditional copyright principles like substantial similarity. Liability is complex: should it fall on the AI developer, the platform provider, or the user who prompted the output?. Professor Qian suggested strict liability for AI producers and fault-based liability (with notice-and-takedown) for platforms, referencing a Chinese case where a platform was required to block infringing prompts.

• Labeling & Traceability: The need for clear labeling of AI-generated content was widely acknowledged for transparency and to combat unfair competition. Deezer discussed developing tools to detect and potentially label AI-generated music on its platform, though differentiating 100% AI vs AI-assisted remains challenging [cite: 2717-2721, 2808].

Broader Implications & The Path Forward

• Impact on Creators & Industries: Creators, particularly songwriters and authors, expressed profound concern about AI displacing human creativity, devaluing their work, and undermining their livelihoods [cite: 2774-2780, 2797-2801, 2949]. Aaron Chaturvedi (Songwriters Association of Canada) called for a WIPO study on legislative options.

• Global South Perspective: Representatives from Cameroon, India, and Brazil highlighted specific challenges for developing countries, including infrastructure gaps, lack of local language datasets, data sovereignty concerns, and the need for frameworks supporting local creative economies [cite: 2023-2029, 2085-2092, 2194-2199]. They stressed the importance of international cooperation and ensuring the Global South isn't merely a consumer of AI developed elsewhere.

• Jurisdictional Challenges & Harmonization: The territorial nature of copyright clashes with the borderless operation of AI [cite: 1842, 1843, 2274-2277]. The risk of legal fragmentation due to differing national approaches was a key concern raised by DDG Sylvie Forbin and others. Many participants stressed the need for continued international dialogue, potentially facilitated by WIPO, to share experiences and explore harmonized solutions or best practices [cite: 1841, 1995, 2116, 2117, 2198, 2199, 2215-2217, 2996]. WIPO announced an upcoming platform ('AI triple I') focused on AI infrastructure issues.

For Scholarly Publishers:

The discussions underscore the urgency for publishers to:

1. Understand the Legal Landscape: Stay informed about evolving national and regional regulations (like the EU AI Act, TDM exceptions, fair use litigation in the US, and emerging frameworks in Asia and the Global South) regarding AI training data and outputs.

2. Review Content Licensing: Assess how existing licenses address (or fail to address) the use of content for AI training. Consider strategies for expressing reservations (opt-outs) or engaging in licensing negotiations for AI use, potentially exploring collective solutions.

3. Address AI-Generated Content: Develop policies regarding the submission and publication of AI-assisted or AI-generated scholarly works, considering authorship criteria, originality standards, and disclosure/labeling requirements.

4. Engage in Dialogue: Participate in industry discussions and contribute to policy development nationally and internationally to ensure the scholarly publishing perspective is heard and a balanced ecosystem is maintained.

The SCCR 46 session made it clear that while AI offers immense potential, navigating its intersection with copyright requires careful consideration, ongoing dialogue, and potentially new global norms or understandings. The conversation has only just begun.

AI, Copyright, and the Future of Scholarly Publishing: Key Takeaways from WIPO’s Latest Session

by Perplexity

What Publishers Need to Know—and Do—Next

The recent WIPO Standing Committee on Copyright and Related Rights (SCCR) session delved into critical issues at the intersection of AI, intellectual property, and global content ecosystems. For scholarly publishers, the discussions revealed both challenges and opportunities. Here’s a breakdown of the top concerns and a roadmap for action.

Key Topics Impacting Scholarly Publishers

1. Generative AI and Copyright Ownership

   • The Dilemma: Who owns AI-generated content—the developer, the user, or no one? The session highlighted unresolved debates about whether AI outputs deserve copyright protection.

   • Publisher Risk: Unclear ownership frameworks could destabilize traditional authorship models, complicating rights management for journals and academic works.

2. AI Training and Copyright Infringement

   • The Debate: Using copyrighted materials to train AI models (e.g., scraping journal articles) risks infringement claims. Publishers emphasized the need for explicit licensing frameworks.

   • Urgency: Unauthorized AI training threatens revenue streams and undermines control over proprietary content.

3. Exceptions for Education and Research

   • Pressure for Broader Access: Calls for expanded limitations and exceptions (L&Es) to support education and research could weaken paywalls.

   • Balancing Act: Publishers must advocate for exceptions that don’t erode subscription models while supporting open science initiatives.

4. Protecting “Signals” in the Digital Age

   • While the Broadcasting Treaty focuses on traditional media, its signal-based approach signals WIPO’s broader push to modernize IP frameworks. Scholarly publishers should watch for analogous rules affecting digital content distribution (e.g., preprint servers, institutional repositories).

5. Transparency and Attribution

   • AI’s “black box” nature raises concerns about proper attribution. Publishers need tools to detect AI-generated content and ensure scholarly integrity.

Strategic Actions for Scholarly Publishers

To navigate this shifting landscape, publishers must act decisively:

1. Advocate for Clear AI Copyright Policies

   • Engage with policymakers at WIPO and national levels to shape regulations that:

     • Clarify ownership of AI-generated works.

     • Require licenses for using copyrighted materials in AI training.

   • Example: Push for a “opt-in” model where publishers must consent to content being used for AI training.

2. Invest in AI Detection and Licensing Tech

   • Develop tools to identify AI-generated submissions and unauthorized use of proprietary content.

   • Explore blockchain or watermarking solutions for traceable attribution.

3. Redefine Licensing Models

   • Create AI-specific licenses for datasets and journal content, akin to Creative Commons for machine learning.

   • Partner with AI developers to negotiate fair compensation structures.

4. Strengthen Open Access Frameworks

   • Proactively design open access policies that address AI use cases (e.g., permitting text mining for non-commercial research while restricting commercial AI training).

5. Collaborate Globally

   • Join coalitions like the Coalition for Responsible Sharing or STM Association to amplify advocacy efforts.

   • Monitor treaties like the Broadcasting Treaty for precedents that could influence digital content rules.

The Bottom Line

AI is reshaping publishing faster than regulations can keep up. Scholarly publishers must pivot from passive observers to active architects of the new IP landscape. By advocating for clarity, investing in adaptive technologies, and reimagining licensing, they can protect their assets while fostering innovation.