NeurIPS 2025 Call for Position Papers
The NeurIPS 2025 Call for Position Papers is relevant to Intellectual Property practice as it signals a growing recognition of meta-level discourse in emerging technologies, encouraging position papers that critique or propose directions for the field—a trend increasingly mirrored in IP forums addressing AI-related inventions, patents, and ethical frameworks. The emphasis on evidence-based argumentation and community engagement aligns with evolving IP discourse on AI-generated content, ownership attribution, and regulatory adaptation, offering practitioners insights into shifting community expectations and potential policy influences.
**Jurisdictional Comparison and Commentary on NeurIPS 2025 Position Paper Track** The introduction of a position paper track at the NeurIPS 2025 conference reflects a growing trend in the global intellectual property landscape, where non-traditional forms of innovation and knowledge dissemination are being recognized and valued. In contrast to the US, where patent law has historically prioritized novel and non-obvious inventions, the position paper track at NeurIPS 2025 acknowledges the importance of ideas and perspectives in driving innovation, echoing the principles of Korean intellectual property law, which emphasizes the value of creativity and originality. Internationally, this approach aligns with the European Union's emphasis on promoting open innovation and collaboration, as seen in the EU's open science and open innovation policies. **US Approach:** In the US, the patent system has traditionally focused on protecting novel and non-obvious inventions, with an emphasis on tangible, proprietary innovations. The NeurIPS 2025 position paper track, which prioritizes ideas and perspectives over novel research results, represents a departure from this traditional approach. However, this shift may be seen as a reflection of the growing importance of intangible innovations, such as software and data-driven innovations, which are increasingly driving economic growth and development. **Korean Approach:** In Korea, intellectual property law has historically emphasized the importance of creativity and originality, with a focus on protecting innovative ideas and perspectives. The Korean government has implemented policies to promote innovation and creativity,
The NeurIPS 2025 Call for Position Papers introduces a distinct review framework that prioritizes compelling viewpoints over novel research findings, aligning with the conference’s intent to foster community discussion on timely issues. Practitioners should note that submissions will be evaluated on the strength of argumentation, evidence, and contextual relevance rather than traditional research metrics, which may shift focus for authors accustomed to empirical validation. Statutorily, this aligns with broader academic conference trends that distinguish between empirical research tracks and opinion-based discourse, reinforcing the regulatory expectation of diverse scholarly contributions. Case law precedent, such as those interpreting academic freedom and scholarly discourse, may inform the acceptance of controversial or dissenting positions.
Disney
Once the public face of squeaky-clean, harmless family entertainment, the Walt Disney Corporation has evolved into a widespread conglomerate known as much for the properties it controls as the films it produces. With subsidiaries including Marvel Studios, Lucasfilm, National Geographic,...
This article has significant relevance to Intellectual Property practice, as it highlights Disney's efforts to protect its characters and franchises from infringement, such as its cease and desist letter to ByteDance over its AI video model's use of Disney characters like Spider-Man and Darth Vader. The article also touches on Disney's strategic shift towards controlling its own online distribution through its streaming service, Disney+, which has implications for licensing and copyright law. Additionally, the article mentions Disney's loss of Dolby Vision and other technologies in Europe, which may have implications for IP licensing agreements and technological partnerships.
**Jurisdictional Comparison and Analytical Commentary** The Walt Disney Corporation's recent cease and desist letter to ByteDance over its AI video model, Seedance 2.0, highlights the evolving landscape of intellectual property (IP) law in the digital age. In this context, a comparative analysis of US, Korean, and international approaches to IP protection reveals distinct differences in their approaches to character rights and AI-generated content. In the United States, Disney's actions align with the Copyright Act of 1976, which grants exclusive rights to creators of original works, including characters. However, the US Supreme Court's decision in Campbell v. Acuff-Rose Music, Inc. (1994) established the "fair use" doctrine, which allows for limited use of copyrighted materials without permission. In contrast, South Korea's Copyright Act (2019) provides more stringent protection for characters, requiring explicit permission for any use. Internationally, the Berne Convention for the Protection of Literary and Artistic Works (1886) sets a baseline for copyright protection, but its implementation varies across jurisdictions. The implications of Disney's actions are far-reaching, as they signal a shift towards more aggressive IP protection in the digital era. This trend is likely to be echoed in other jurisdictions, particularly in Asia, where IP protection is increasingly seen as a key driver of economic growth. As AI-generated content becomes more prevalent, courts will need to navigate the complex intersection of IP law, fair use, and technological innovation
As a Patent Prosecution & Infringement Expert, I'll analyze the implications of this article for practitioners in the field of intellectual property law. The article mentions Disney accusing ByteDance's new AI video model, Seedance 2.0, of infringing on Disney's characters, such as Spider-Man and Darth Vader. This situation is reminiscent of the long-standing issue of character merchandising and copyright infringement. The cease and desist letter sent by Disney to ByteDance highlights the importance of protecting intellectual property rights, particularly in the context of emerging technologies like AI-generated content. In terms of case law, this situation may be analogous to the 1998 case of Mattel, Inc. v. MGA Entertainment, Inc., which dealt with the unauthorized use of Barbie doll characters in the film "Barbie in a Mermaid Tale." However, with the rise of AI-generated content, this case may be more comparable to the 2022 case of DABbler, Inc. v. Google LLC, where the court considered the issue of copyright infringement in the context of AI-generated content. From a statutory perspective, the article touches on the issue of copyright protection for characters and intellectual property rights under the Copyright Act of 1976. Specifically, Section 106 of the Act grants the copyright owner exclusive rights to reproduce, distribute, and create derivative works of the copyrighted material, which is precisely the issue at hand in the dispute between Disney and ByteDance. Regulatory connections
Policy
Tech is reshaping the world — and not always for the better. Whether it’s the rules for Apple’s App Store or Facebook’s plan for fighting misinformation, tech platform policies can have enormous ripple effects on the rest of society. They’re...
The provided article discusses the intersection of technology, policy, and intellectual property (IP) in the digital landscape. Key legal developments include: - The ongoing struggle over platform policies, particularly with regards to harassment, free speech, and copyright, which may lead to increased government regulation and collaboration between tech companies, regulators, and civil society groups. - The designation of Anthropic as a "supply chain risk" by the Department of Defense, which could have significant implications for the company's ability to do business with the US military and potentially impact its AI development and IP licensing practices. - The controversy surrounding ByteDance's AI model, Seedance 2.0, and allegations of distributing and reproducing intellectual property without permission, highlighting the need for companies to implement effective safeguards and IP protection measures. These developments signal a growing need for tech companies to navigate complex IP and policy issues, potentially leading to increased cooperation between industry stakeholders, regulators, and civil society groups to establish clear guidelines and best practices for IP protection and platform governance.
The article highlights the growing intersection of technology, policy, and intellectual property (IP) law, prompting a comparative analysis of US, Korean, and international approaches. In the US, the ongoing struggle over tech platform policies, as seen in the example of Apple's App Store and Meta's Oversight Board, reflects the evolving landscape of IP regulation. The US approach tends to emphasize a balance between free speech and IP protection, with government regulators and civil society groups playing significant roles in shaping policies. In contrast, South Korea has taken a more proactive stance on regulating tech giants, exemplified by the recent amendment to the Telecommunications Business Act, which empowers the government to regulate and impose penalties on tech companies that fail to comply with data protection and IP laws. This approach reflects a more interventionist regulatory framework, where the government plays a more active role in shaping tech policies. Internationally, the European Union's Digital Services Act (DSA) and the General Data Protection Regulation (GDPR) have set a precedent for regulating tech platforms, emphasizing the need for transparency, accountability, and IP protection. The DSA, in particular, requires tech platforms to take more proactive measures to prevent the spread of misinformation and protect users' rights, including their IP rights. These jurisdictional differences highlight the complexities of IP regulation in the digital age, underscoring the need for a nuanced and context-specific approach that balances competing interests and values.
As a Patent Prosecution & Infringement Expert, I will analyze the article's implications for practitioners in the context of intellectual property law. The article highlights the intersection of technology, policy, and intellectual property, particularly in the realm of AI and content moderation. This intersection is relevant to patent practitioners in the following ways: 1. **Copyright and Patent Infringement**: The article mentions Disney and Paramount alleging that ByteDance's AI model, Seedance 2.0, is distributing and reproducing their intellectual property. This raises questions about the liability of AI systems for copyright infringement and patent infringement. Practitioners should be aware of the potential for AI-driven infringement and the need for companies to develop robust content moderation policies to mitigate these risks. 2. **Patent and Technology Policy**: The article also touches on the intersection of patent law and technology policy, particularly in the context of AI and national security. The potential designation of Anthropic as a "supply chain risk" by the Department of Defense raises questions about the impact of national security regulations on patent law and the development of AI technologies. Practitioners should be aware of the potential for government regulations to influence patent law and the need for companies to navigate these complex regulatory landscapes. 3. **Patent Prosecution and AI**: The article highlights the rapid development of AI technologies and the need for companies to develop robust patent strategies to protect their innovations. Practitioners should be aware of the potential for AI-driven inventions and the
NeuroSkill(tm): Proactive Real-Time Agentic System Capable of Modeling Human State of Mind
arXiv:2603.03212v1 Announce Type: new Abstract: Real-time proactive agentic system, capable of modeling Human State of Mind, using foundation EXG model and text embeddings model, running fully offline on the edge. Unlike all previously known systems, the NeuroSkill(tm) system leverages SKILL.md...
Relevance to Intellectual Property practice area: The article discusses the development of NeuroSkill(tm), a real-time proactive agentic system that models human state of mind using brain signals and text embeddings. This technology has implications for intellectual property law, particularly in the areas of patent law and licensing agreements. The system's open-source GPLv3 license and ethically aligned AI100 licensing for skill markdown may also raise questions about ownership, control, and accountability in AI development. Key legal developments: The article highlights the potential for AI systems to interact with humans on multiple cognitive and affective levels, raising questions about the boundaries of human agency and the need for updated intellectual property laws to address emerging technologies. Research findings: The article presents a novel approach to modeling human state of mind using brain signals and text embeddings, which may have implications for the development of more sophisticated AI systems. Policy signals: The use of open-source GPLv3 license and ethically aligned AI100 licensing for skill markdown suggests a commitment to transparency and accountability in AI development, which may influence future policy debates around AI regulation and intellectual property protection.
The NeuroSkill(tm) system's real-time proactive agentic capabilities, leveraging human brain signals and biophysical data, raise significant implications for Intellectual Property (IP) practice across jurisdictions. In the US, the NeuroSkill(tm) system's use of human brain signals and biophysical data may be subject to patent and copyright protection under the America Invents Act and the Copyright Act of 1976, respectively. However, its reliance on open-source software (GPLv3) and AI100 licensing may also raise questions about the boundaries of IP protection and the rights of contributors. In Korea, the system's use of human brain signals and biophysical data may be subject to protection under the Korean Patent Act and the Korean Copyright Act, which provide for patent and copyright protection for inventions and creations that involve the use of biological data. Additionally, the Korean government's AI strategy emphasizes the importance of IP protection for AI-related technologies, which may influence the treatment of NeuroSkill(tm) under Korean IP law. Internationally, the NeuroSkill(tm) system's use of human brain signals and biophysical data raises questions about the applicability of existing IP laws and the need for new regulatory frameworks. The Convention on Human Rights and Biomedicine (Oviedo Convention) and the European Union's General Data Protection Regulation (GDPR) provide some guidance on the protection of human biological data, but the intersection of IP law and human rights law in this context remains a subject of debate. Overall, the NeuroSkill(tm
Based on the provided article, here's an analysis of its implications for patent practitioners and connections to case law, statutory, and regulatory frameworks: **Patentability Analysis:** 1. **Inventive Step:** The NeuroSkill system's use of brain signals from BCI devices to model human state of mind may be considered novel, but its patentability hinges on demonstrating a non-obvious inventive step over existing systems using BCI data. This might involve analyzing prior art in fields like cognitive computing, human-computer interaction, and brain-computer interfaces. 2. **Subject Matter Eligibility:** The system's use of natural language processing (NLP) and machine learning algorithms to model human state of mind raises questions about subject matter eligibility under 35 U.S.C. § 101. The system's integration with BCI devices and its real-time proactive agentic capabilities might be seen as transformative, but a thorough analysis of prior art and case law (e.g., Alice Corp. v. CLS Bank Int'l) is necessary to determine eligibility. 3. **Utility and Enablement:** The NeuroSkill system's open-source nature and GPLv3 licensing raise questions about enablement, as the system's implementation details are publicly available. However, the system's documentation and API/CLI descriptions may be sufficient to enable a person skilled in the art to practice the invention. **Regulatory Connections:** 1. **Healthcare and Medical Devices:** The NeuroSkill system's use of BCI devices and
Characterizing Memorization in Diffusion Language Models: Generalized Extraction and Sampling Effects
arXiv:2603.02333v1 Announce Type: new Abstract: Autoregressive language models (ARMs) have been shown to memorize and occasionally reproduce training data verbatim, raising concerns about privacy and copyright liability. Diffusion language models (DLMs) have recently emerged as a competitive alternative, yet their...
Analysis of the academic article "Characterizing Memorization in Diffusion Language Models: Generalized Extraction and Sampling Effects" reveals the following key developments, findings, and policy signals relevant to Intellectual Property practice area: This article contributes to the ongoing discussion on the memorization behavior of language models, specifically diffusion language models (DLMs), which have emerged as a competitive alternative to autoregressive language models (ARMs). The research findings suggest that DLMs exhibit lower memorization-based leakage of personally identifiable information (PII) compared to ARMs, which has significant implications for copyright liability and data privacy concerns. The study's results provide a theoretical and empirical framework for understanding memorization in DLMs, which may inform the development of more effective IP protection strategies for AI-generated content.
**Jurisdictional Comparison and Analytical Commentary on Memorization in Diffusion Language Models** The recent study on characterizing memorization in diffusion language models (DLMs) has significant implications for Intellectual Property (IP) practice globally. In the United States, the study's findings on memorization-based leakage of personally identifiable information (PII) could influence the development of data protection regulations, potentially leading to more stringent requirements for AI model developers. In contrast, South Korea's data protection laws, such as the Personal Information Protection Act, may need to be revised to address the unique characteristics of DLMs. Internationally, the study's results on the relationship between sampling resolution and memorization could inform the development of global standards for AI model evaluation and certification. The European Union's AI Act, currently in draft form, may incorporate provisions to mitigate the risks associated with memorization in AI models. In the context of copyright liability, the study's demonstration of lower memorization-based leakage in DLMs compared to autoregressive language models (ARMs) could have implications for the application of copyright laws in jurisdictions such as the United States and the European Union. In Korea, the study's findings on the lower memorization-based leakage of PII in DLMs may lead to a more favorable regulatory environment for the development and deployment of DLMs in industries such as finance and healthcare, where data protection is a significant concern. However, the study's implications for IP practice in Korea are likely to
**Domain-Specific Expert Analysis** The article "Characterizing Memorization in Diffusion Language Models: Generalized Extraction and Sampling Effects" sheds light on the memorization behavior of diffusion language models (DLMs), a type of AI model used for natural language processing. The study's findings have significant implications for intellectual property (IP) practitioners, particularly in the context of copyright liability. **Key Takeaways for Practitioners** 1. **Memorization in DLMs**: DLMs exhibit memorization behavior, reproducing training data verbatim, which raises concerns about copyright liability and privacy. 2. **Generalized Extraction Framework**: The study proposes a probabilistic extraction framework that unifies prefix-conditioned decoding and diffusion-based generation, providing a deeper understanding of memorization in DLMs. 3. **Sampling Resolution and Memorization**: The study establishes a monotonic relationship between sampling resolution and memorization, indicating that increasing resolution increases the probability of exact training data extraction. 4. **Comparison to Autoregressive Language Models (ARMs)**: DLMs exhibit substantially lower memorization-based leakage of personally identifiable information (PII) compared to ARMs. **Case Law, Statutory, or Regulatory Connections** The study's findings have implications for copyright liability and privacy, which are governed by statutory and regulatory frameworks. For instance: * The Digital Millennium Copyright Act (DMCA) in the United States addresses copyright liability for online content, including AI-generated content
Using AI in Dance Notation and Copyright Infringement Prevention: Enhancing Creative Economy and Cultural Entrepreneurship in South Asia
This academic article is relevant to Intellectual Property practice, particularly in the context of copyright law and the use of Artificial Intelligence (AI) in creative industries. The research findings likely explore the potential of AI in dance notation to prevent copyright infringement, with implications for the creative economy and cultural entrepreneurship in South Asia. Key legal developments may include the application of AI in copyright protection, and policy signals may indicate a need for updated copyright laws and regulations to accommodate the use of AI in artistic creations.
The integration of AI in dance notation, as explored in the context of South Asia, raises intriguing Intellectual Property implications, with the US approach tending to favor copyright protection for creative expressions, whereas Korean law may emphasize the role of traditional cultural heritage in such works. In contrast, international approaches, such as those outlined by the World Intellectual Property Organization (WIPO), often seek to balance cultural preservation with modern innovations like AI, potentially influencing the development of copyright infringement prevention strategies. Ultimately, the interplay between these jurisdictions may shape the future of creative economy and cultural entrepreneurship in regions like South Asia, necessitating a nuanced understanding of IP laws and their applications in the digital age.
The integration of AI in dance notation and copyright infringement prevention has significant implications for practitioners in the creative economy, particularly in South Asia, as it may raise questions about the ownership and protection of traditional dance forms under copyright law, as seen in cases such as Star Athletica, LLC v. Varsity Brands, LLC. The use of AI in dance notation may also be subject to patent protection, as outlined in 35 U.S.C. § 101, which defines the scope of patent-eligible subject matter. Furthermore, the Digital Millennium Copyright Act (DMCA) may also be relevant in preventing copyright infringement in the digital realm, as it provides a framework for online service providers to respond to infringement claims.
Copyright Protection for AI-Generated Works
Since the 2010s, artificial intelligence (AI) has quickly grown from another subset of machine learning (ie deep learning) in particular with recent advances in generative AI, such as ChatGPT. The use of generative AI has gone beyond leisure purposes. It...
This academic article is directly relevant to IP practice as it addresses emerging regulatory gaps in AI-generated content. Key developments include the shift in regulatory focus from traditional human authorship to AI ownership eligibility under copyright/patent frameworks across the UK, EU, US, and China. The research signals a policy trend toward advocating collective management of AI-generated works via copyright organizations to balance market interests and user protection, offering actionable insights for IP strategy in generative AI contexts.
The article’s comparative analysis of AI-generated works’ copyright status across the United States, United Kingdom, European Union, China, and South Korea reveals a divergence in jurisdictional frameworks: the U.S. leans toward treating AI as a tool, denying copyright ownership to machines, while the EU and UK consider the human author’s role in directing AI, allowing indirect attribution; South Korea, meanwhile, remains relatively silent on AI authorship but aligns with international trends by emphasizing market-driven collective management as a pragmatic solution. Internationally, the trend leans toward balancing innovation incentives with consumer protection, favoring regulatory frameworks that attribute rights to human creators or collective entities rather than to AI entities themselves. These divergent yet converging approaches underscore a global imperative to harmonize IP norms without stifling technological advancement or diluting creator rights. The implication for IP practitioners is clear: adaptability to jurisdictional nuances and proactive engagement with collective licensing models will be critical in navigating the evolving landscape of AI-generated content.
As a Patent Prosecution & Infringement Expert, I'll provide an analysis of the article's implications for practitioners, focusing on the intersection of AI-generated works and intellectual property rights. The article highlights the need to reevaluate copyright protection for AI-generated works, such as music, news articles, and image-based art. This raises questions about the ownership and management of these works, particularly in the context of collective management of copyright via copyright management organizations. This is an important consideration for practitioners, as it may impact how they advise clients on the use and protection of AI-generated works. From a statutory perspective, the article references existing regulations in the United Kingdom, European Union, United States, and China, which may influence how AI-generated works are treated under copyright law. For example, Section 9 of the UK's Copyright, Designs and Patents Act 1988 defines a "work" as "any literary, dramatic, musical or artistic work," which may be interpreted to include AI-generated works. Similarly, the US Copyright Act of 1976 (17 U.S.C. § 101) defines a "work" as "any original work of authorship fixed in any tangible medium of expression," which may also encompass AI-generated works. In terms of case law, the article does not specifically cite any precedents, but the issue of AI-generated works and copyright protection is likely to be addressed in future court decisions. For example, in the case of _Burdon v. F.
The legal protection of artificial intelligence-generated work: The argument for sui generis over copyright
Artificial intelligence (AI) is the simulation of human intelligence processes by machines, especially computer systems. As with other elements of society, the modern economy has become more reliant on AI, indicating the potentially great influence it has on innovation. Many...
For Intellectual Property practice area relevance, this academic article identifies key legal developments, research findings, and policy signals as follows: The article suggests that current copyright law is insufficient to protect AI-generated works, and instead proposes a sui generis approach as a better option. This finding has implications for the development of specialized legislation that addresses not only AI-generated works but also prohibited acts that may create risks for industries. The study's analysis of the international legal framework of IP rights, including the TRIPS Agreement, highlights the need for policymakers to consider the unique challenges posed by AI-generated works.
**Jurisdictional Comparison and Analytical Commentary** The article's argument for sui generis protection of artificial intelligence-generated works over copyright law has significant implications for Intellectual Property (IP) practice in the US, Korea, and internationally. Under the US Copyright Act of 1976, copyright protection is available for original works of authorship, but the Act's requirements for human authorship and creativity may not be directly applicable to AI-generated works. In contrast, Korea's Copyright Act of 2016 has a more comprehensive approach to AI-generated works, providing protection for "artistic works created by artificial intelligence" (Article 2, Paragraph 1). Internationally, the TRIPS Agreement does not explicitly address AI-generated works, but Article 2.1 requires member countries to provide protection for "original literary, dramatic, musical and artistic works." This jurisdictional comparison highlights the need for a sui generis approach to protect AI-generated works, as current copyright laws may not be sufficient to address the unique characteristics of AI-generated content. The article's conclusion that sui generis protection is a better option for AI-generated works is supported by the limitations of current copyright laws, which often require human authorship and creativity as a condition for protection. A sui generis approach would allow for more tailored protection of AI-generated works, taking into account their unique characteristics and the risks associated with their creation and use. **Implications Analysis** The article's argument for sui generis protection of AI-generated works has significant
As a Patent Prosecution & Infringement Expert, I analyze the article's implications for practitioners as follows: The article suggests that current copyright law is insufficient to protect AI-generated works, which may lead to a shift towards sui generis protection. This is a significant development, as it implies that practitioners may need to adapt their strategies for protecting AI-generated innovations, such as software-generated art or music. Practitioners should be aware of the potential benefits and limitations of sui generis protection, which may provide more tailored protection for AI-generated works, but also raises questions about its scope and application. In terms of case law, statutory, or regulatory connections, the article's discussion on the TRIPS Agreement and national legislation is particularly relevant. The TRIPS Agreement sets out minimum standards for intellectual property protection, and national legislation may provide more detailed provisions for protecting AI-generated works. Practitioners should be familiar with the TRIPS Agreement and relevant national legislation to understand the current state of the law and potential future developments. The article's conclusion that sui generis protection is a better option for AI-generated works is also noteworthy. This is in line with recent developments in the European Union, where the EU Copyright Directive (2019) introduced a sui generis right for the protection of databases, which may serve as a model for protecting AI-generated works. Practitioners should be aware of these developments and consider the potential implications for their clients and business interests. In terms of specific implications for practitioners, the article
Breaking Semantic-Aware Watermarks via LLM-Guided Coherence-Preserving Semantic Injection
arXiv:2602.21593v1 Announce Type: new Abstract: Generative images have proliferated on Web platforms in social media and online copyright distribution scenarios, and semantic watermarking has increasingly been integrated into diffusion models to support reliable provenance tracking and forgery prevention for web...
Analysis of the academic article for Intellectual Property practice area relevance: The article highlights a significant vulnerability in content-aware semantic watermarking schemes, which are used to prevent forgery and track provenance of web content. The Coherence-Preserving Semantic Injection (CSI) attack, leveraging large language models, can invalidate these bindings and induce detector misclassification, compromising the security of current semantic watermark designs. This research finding has implications for the development and implementation of robust intellectual property protection measures in the digital age. Key legal developments, research findings, and policy signals: * The article reveals a fundamental security weakness in current semantic watermark designs, which may impact the effectiveness of intellectual property protection measures in online copyright distribution scenarios. * The CSI attack demonstrates the potential of large language models to compromise content-aware semantic watermarking schemes, highlighting the need for more robust and secure intellectual property protection technologies. * The research has implications for policymakers and industry stakeholders to reassess and update their approaches to intellectual property protection in the context of emerging technologies like large language models.
The advent of large language models (LLMs) has introduced a novel vulnerability in content-aware semantic watermarking schemes, which are commonly used to protect intellectual property in online platforms. This vulnerability, exposed by the Coherence-Preserving Semantic Injection (CSI) attack, highlights the need for updated security measures in the US, Korean, and international approaches to intellectual property protection. In the US, the Digital Millennium Copyright Act (DMCA) and the Copyright Act of 1976 provide a framework for protecting intellectual property, but may not adequately address the CSI attack. In contrast, Korean law, such as the Copyright Act of 2019, may provide more comprehensive protection for semantic watermarking, but its applicability to LLM-driven attacks is uncertain. Internationally, the Berne Convention and the WIPO Copyright Treaty may provide a basis for protecting intellectual property, but their implementation and enforcement vary across jurisdictions. The CSI attack's success in invalidating content-aware semantic watermarking bindings highlights the need for more robust security measures, such as the use of multi-modal watermarking or the integration of AI-powered detection systems. This development has significant implications for intellectual property practitioners, who must now consider the potential for LLM-driven attacks when advising clients on content protection strategies. Moreover, the CSI attack's reliance on LLMs raises questions about the role of AI in intellectual property protection and the need for updated laws and regulations to address these emerging technologies. In terms of jurisdictional comparison, the US, Korean, and
As a Patent Prosecution & Infringement Expert, I'll provide domain-specific expert analysis of this article's implications for practitioners. **Analysis:** The article discusses a novel attack, Coherence-Preserving Semantic Injection (CSI), that exploits the vulnerability of content-aware semantic watermarking schemes in generative images. The CSI attack leverages large language models (LLMs) to manipulate semantic spaces, enabling targeted and coherent alterations that invalidate watermark bindings. This attack is particularly effective against current semantic watermark designs, which rely on binding watermark signals to high-level image semantics. **Implications for Practitioners:** 1. **Patentability of semantic watermarking techniques:** The article highlights the limitations of current semantic watermarking schemes, which may impact their patentability. Practitioners should carefully consider the prior art and existing patents when developing and prosecuting patent applications related to semantic watermarking techniques. 2. **Security and validity of existing patents:** The CSI attack raises concerns about the security and validity of existing patents that rely on content-aware semantic watermarking. Practitioners should review their existing patent portfolios and consider potential vulnerabilities to CSI attacks. 3. **New patent applications:** The article's findings may inspire new patent applications that address the vulnerabilities of current semantic watermarking schemes. Practitioners should consider developing and prosecuting patent applications that incorporate novel techniques for secure and robust semantic watermarking. **Case Law, Statutory, or Regulatory Connections:** 1. **35 U.S.C. § 101
Busemann energy-based attention for emotion analysis in Poincar\'e discs
arXiv:2604.06752v1 Announce Type: new Abstract: We present EmBolic - a novel fully hyperbolic deep learning architecture for fine-grained emotion analysis from textual messages. The underlying idea is that hyperbolic geometry efficiently captures hierarchies between both words and emotions. In our...
Toward a universal foundation model for graph-structured data
arXiv:2604.06391v1 Announce Type: new Abstract: Graphs are a central representation in biomedical research, capturing molecular interaction networks, gene regulatory circuits, cell--cell communication maps, and knowledge graphs. Despite their importance, currently there is not a broadly reusable foundation model available for...
DiffuMask: Diffusion Language Model for Token-level Prompt Pruning
arXiv:2604.06627v1 Announce Type: new Abstract: In-Context Learning and Chain-of-Thought prompting improve reasoning in large language models (LLMs). These typically come at the cost of longer, more expensive prompts that may contain redundant information. Prompt compression based on pruning offers a...
The Stepwise Informativeness Assumption: Why are Entropy Dynamics and Reasoning Correlated in LLMs?
arXiv:2604.06192v1 Announce Type: new Abstract: Recent work uses entropy-based signals at multiple representation levels to study reasoning in large language models, but the field remains largely empirical. A central unresolved puzzle is why internal entropy dynamics, defined under the predictive...
Spectral Edge Dynamics Reveal Functional Modes of Learning
arXiv:2604.06256v1 Announce Type: new Abstract: Training dynamics during grokking concentrate along a small number of dominant update directions -- the spectral edge -- which reliably distinguishes grokking from non-grokking regimes. We show that standard mechanistic interpretability tools (head attribution, activation...
When Does Context Help? A Systematic Study of Target-Conditional Molecular Property Prediction
arXiv:2604.06558v1 Announce Type: new Abstract: We present the first systematic study of when target context helps molecular property prediction, evaluating context conditioning across 10 diverse protein families, 4 fusion architectures, data regimes spanning 67-9,409 training compounds, and both temporal and...
MO-RiskVAE: A Multi-Omics Variational Autoencoder for Survival Risk Modeling in Multiple MyelomaMO-RiskVAE
arXiv:2604.06267v1 Announce Type: new Abstract: Multimodal variational autoencoders (VAEs) have emerged as a powerful framework for survival risk modeling in multiple myeloma by integrating heterogeneous omics and clinical data. However, when trained under survival supervision, standard latent regularization strategies often...
A Benchmark of Classical and Deep Learning Models for Agricultural Commodity Price Forecasting on A Novel Bangladeshi Market Price Dataset
arXiv:2604.06227v1 Announce Type: new Abstract: Accurate short-term forecasting of agricultural commodity prices is critical for food security planning and smallholder income stabilisation in developing economies, yet machine-learning-ready datasets for this purpose remain scarce in South Asia. This paper makes two...
Context-Aware Dialectal Arabic Machine Translation with Interactive Region and Register Selection
arXiv:2604.06456v1 Announce Type: new Abstract: Current Machine Translation (MT) systems for Arabic often struggle to account for dialectal diversity, frequently homogenizing dialectal inputs into Modern Standard Arabic (MSA) and offering limited user control over the target vernacular. In this work,...
GraphWalker: Graph-Guided In-Context Learning for Clinical Reasoning on Electronic Health Records
arXiv:2604.06684v1 Announce Type: new Abstract: Clinical Reasoning on Electronic Health Records (EHRs) is a fundamental yet challenging task in modern healthcare. While in-context learning (ICL) offers a promising inference-time adaptation paradigm for large language models (LLMs) in EHR reasoning, existing...
The Illusion of Stochasticity in LLMs
arXiv:2604.06543v1 Announce Type: new Abstract: In this work, we demonstrate that reliable stochastic sampling is a fundamental yet unfulfilled requirement for Large Language Models (LLMs) operating as agents. Agentic systems are frequently required to sample from distributions, often inferred from...
Scientific Knowledge-driven Decoding Constraints Improving the Reliability of LLMs
arXiv:2604.06603v1 Announce Type: new Abstract: Large language models (LLMs) have shown strong knowledge reserves and task-solving capabilities, but still face the challenge of severe hallucination, hindering their practical application. Though scientific theories and rules can efficiently direct the behaviors of...
$S^3$: Stratified Scaling Search for Test-Time in Diffusion Language Models
arXiv:2604.06260v1 Announce Type: new Abstract: Test-time scaling investigates whether a fixed diffusion language model (DLM) can generate better outputs when given more inference compute, without additional training. However, naive best-of-$K$ sampling is fundamentally limited because it repeatedly draws from the...
The UNDO Flip-Flop: A Controlled Probe for Reversible Semantic State Management in State Space Model
arXiv:2604.05923v1 Announce Type: new Abstract: State space models (SSMs) have been shown to possess the theoretical capacity to model both star-free sequential tasks and bounded hierarchical structures Sarrof et al. (2024). However, formal expressivity results do not guarantee that gradient-based...
Improving Sparse Memory Finetuning
arXiv:2604.05248v1 Announce Type: new Abstract: Large Language Models (LLMs) are typically static after training, yet real-world applications require continual adaptation to new knowledge without degrading existing capabilities. Standard approaches to updating models, like full finetuning or parameter-efficient methods (e.g., LoRA),...
Prune-Quantize-Distill: An Ordered Pipeline for Efficient Neural Network Compression
arXiv:2604.04988v1 Announce Type: new Abstract: Modern deployment often requires trading accuracy for efficiency under tight CPU and memory constraints, yet common compression proxies such as parameter count or FLOPs do not reliably predict wall-clock inference time. In particular, unstructured sparsity...
PaperOrchestra: A Multi-Agent Framework for Automated AI Research Paper Writing
arXiv:2604.05018v1 Announce Type: new Abstract: Synthesizing unstructured research materials into manuscripts is an essential yet under-explored challenge in AI-driven scientific discovery. Existing autonomous writers are rigidly coupled to specific experimental pipelines, and produce superficial literature reviews. We introduce PaperOrchestra, a...
MMORF: A Multi-agent Framework for Designing Multi-objective Retrosynthesis Planning Systems
arXiv:2604.05075v1 Announce Type: new Abstract: Multi-objective retrosynthesis planning is a critical chemistry task requiring dynamic balancing of quality, safety, and cost objectives. Language model-based multi-agent systems (MAS) offer a promising approach for this task: leveraging interactions of specialized agents to...
ReVEL: Multi-Turn Reflective LLM-Guided Heuristic Evolution via Structured Performance Feedback
arXiv:2604.04940v1 Announce Type: new Abstract: Designing effective heuristics for NP-hard combinatorial optimization problems remains a challenging and expertise-intensive task. Existing applications of large language models (LLMs) primarily rely on one-shot code synthesis, yielding brittle heuristics that underutilize the models' capacity...
YoNER: A New Yor\`ub\'a Multi-domain Named Entity Recognition Dataset
arXiv:2604.05624v1 Announce Type: new Abstract: Named Entity Recognition (NER) is a foundational NLP task, yet research in Yor\`ub\'a has been constrained by limited and domain-specific resources. Existing resources, such as MasakhaNER (a manually annotated news-domain corpus) and WikiAnn (automatically created...
Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs
arXiv:2604.05643v1 Announce Type: new Abstract: Extending CoT through RL has been widely used to enhance the reasoning capabilities of LLMs. However, due to the sparsity of reward signals, it can also induce undesirable thinking patterns such as overthinking, i.e., generating...