New Litigation Front: Shareholder Suits Allege Breach of Fiduciary Duty #
- Shareholders file derivative suits in new wave of copyright-related litigation against AI companies (CCH Incorporated) — A significant new legal strategy has emerged beyond direct copyright claims: shareholder derivative lawsuits. These suits, filed in mid-2026, target the officers and directors of AI companies like Adobe and NVIDIA, alleging they breached their fiduciary duties by knowingly developing AI products using pirated datasets. The complaints, such as SEIU Plan Master Trust v. Narayen and Berliner v. Huang, argue that this strategy exposed the companies to legal liability and reputational damage, causing drops in stock prices and, in one instance, allegedly contributing to a CEO’s departure. This approach shifts the focus from copyright holders suing AI firms to investors suing the corporate decision-makers for allegedly failing to manage risk.
Fair Use Under Scrutiny: Foundational Assumptions Challenged in Court #
- AI Fair Use Rulings Are Built on an Assumption No Court Has Examined (EIN News) — A new amicus brief filed on August 26, 2026, in the In re Mosaic LLM Litigation directly challenges the core analogy underpinning recent fair use victories for AI developers. The AI Rights Project argues that courts in cases like Bartz v. Anthropic have treated machine training as equivalent to “human reading” without examining the validity of that premise. The brief contends that where fair use has historically assumed a person exercising creative judgment, machine-scale training displaces that human-scale assumption. It posits that extending fair use to cover industrial-scale data ingestion is a decision for Congress, not one courts should make by analogy, introducing a novel argument that could reframe the fair use debate.
Market Structure: Licensing Deals Create a Multi-Billion Dollar Market #
- Every AI Content Licensing Deal, Mapped (2023-2026) (LLM Pulse) — The AI content licensing market has rapidly evolved from ad-hoc deals to a structured marketplace with significant financial flows. A comprehensive map, updated August 17, 2026, details dozens of agreements between AI developers (OpenAI, Google, Apple) and content owners. The report quantifies several major deals, including OpenAI’s reported $250 million five-year agreement with News Corp and Reddit’s disclosure of $203 million in aggregate data-licensing contracts with buyers like Google. This data illustrates a clear trend of major AI labs paying for training data, establishing a parallel “license-and-pay” track alongside ongoing “scrape-and-defend” litigation.
Regulatory Paths Diverge as Legislative Proposals Evolve #
- The U.S. and the EU regulate AI differently (Fast Company) — As of late August 2026, the regulatory gap between the US and EU is solidifying, with copyright law becoming a central battleground. The EU’s approach, combining the AI Act with the DSM Directive’s opt-out provision for text and data mining (TDM), creates a system where developers must actively audit training data against a list of rights-holder objections. In contrast, the US continues to rely on the four-factor Fair Use doctrine, leaving the legality of training to be decided through case-by-case litigation. Meanwhile, legislative efforts in the US, such as the proposed “Generative AI Copyright Disclosure Act of 2024” which would have mandated disclosure of training data, have so far failed to pass, cementing the judicial-led approach.
- Assessing the feasibility of collective licensing of in-copyright works as training data for generative AI systems (PNAS) — As an alternative to litigation or direct licensing, collective licensing is gaining traction as a potential solution. Analysis from July 2026 highlights that organizations like the Association of American Publishers (AAP) and the Copyright Clearance Center (CCC) are actively promoting voluntary collective licensing frameworks. The CCC has already developed a specific license for using works as training data. This approach would create a centralized system for clearing rights and distributing royalties, though its feasibility depends heavily on the outcomes of major fair use lawsuits, which could either compel developers to the table or render the need for such licenses moot.
Synthetic Data’s Legal Standing Under EU Law #
- Legal Considerations in the Use of Synthetic Data for AI Development and Finetuning: The Case of LLMs4EU (LREC 2026) — A May 2026 academic analysis explores the complex legal status of synthetic data, a key proposed alternative to using copyrighted works for training. The paper, using the LLMs4EU project as a case study, examines whether synthetic data qualifies as “personal data” under GDPR and assesses the downstream copyright risks. A key issue identified is that if the model used to generate the synthetic data was itself trained on infringing content, the resulting synthetic data may still carry infringement liability. This highlights that synthetic data is not a simple legal “silver bullet” but requires robust governance frameworks to ensure the entire data value chain is compliant with EU regulations.
- Emerging pattern: Shareholder derivative lawsuits represent a new and potent vector for challenging the use of unauthorized training data. This tactic sidesteps the complexities of proving copyright infringement and instead focuses on corporate governance and risk management, potentially creating pressure on boards and executives independent of the outcomes of the core copyright cases.