Art & Culture Law · 2026-06-19

⚖️ Art-Culture-Law Watcher — 2026-06-19

🎭 Senate Judiciary Committee Unanimously Clears S. 4591 to Combat Unauthorized AI replicas
🎻 Musicians Union Sues Major Record Labels Over Unauthorized AI Licensing Settlements
🎹 Suno Battles to Conceal Training Dataset Volume in Court as RIAA Discovery Closes
🏛️ Third Circuit Weighs Fair Use in Westlaw Copyright Appeal Against AI Startup Ross
🇪🇺 European AI Office Releases Final Transparency Code of Practice for Watermarking
📈 Shareholders Sue Adobe Board Over AI Dataset Contamination and Misleading Statements

---

🎭 Senate Judiciary Committee Unanimously Clears S. 4591 to Combat Unauthorized AI replicas

The legislative architecture governing artificial intelligence and intellectual property reached a historic bipartisan milestone on Thursday, June 18, 2026, as the Senate Judiciary Committee unanimously cleared the No Fakes Act. Designated formally as S. 4591, the NO FAKES Act of 2026, this legislation represents the first comprehensive federal effort to establish a statutory federal intellectual property right over an individual’s voice and visual likeness. Co-sponsored by a bipartisan coalition led by Senators Chris Coons and Marsha Blackburn, the bill specifically targets the rapid, unregulated proliferation of unauthorized generative AI deepfakes. By consolidating what has historically been a fragmented patchwork of state-level rights of publicity, S. 4591 seeks to provide a uniform, nationwide shield for digital personas.

A major flashpoint in the legislative negotiations involved the specific liability standards imposed on downstream online platforms. Under the current text of S. 4591, platforms that knowingly host or facilitate the distribution of unauthorized digital replicas can face statutory damages starting at $750,000 per violation. This steep liability ceiling has prompted intense lobbying from both creative guilds and technology associations. As Roll Call reports, Senator Blackburn has been actively spearheading negotiations directly with the White House to finalize the preemption language, ensuring that the federal statute successfully balances individual IP enforcement against broader digital safety initiatives.

The committee’s vote marks a profound shift in how the law conceptualizes the human voice and face—transforming likeness from a tort-protected personal interest into a fully assignable, licensable intellectual property asset. Unlike copyrights, which are typically owned by corporate entities or publishers, this digital replica right remains unassignable during the creator's lifetime, though it can be licensed for commercial use. The Washington Times reports that the bill’s passage to the Senate floor is being closely watched by digital platforms, who argue that the lack of clear preemption clauses could still leave them vulnerable to overlapping state-level claims. Nonetheless, the bill's unanimous advancement out of committee signals that federal regulation of digital likeness is no longer a theoretical debate, but an imminent legal reality.

Sources:

---

🎻 Musicians Union Sues Major Record Labels Over Unauthorized AI Licensing Settlements

A major civil war has erupted within the music industry's executive and labor ranks. On June 5, 2026, the American Federation of Musicians (AFM) sued major record labels—including Universal Music Group, Sony Music Entertainment, and Warner Music Group—in a federal lawsuit that could reshape the economics of AI music generation. The union alleges that the major labels' settlements of copyright disputes with prominent generative AI music startups, such as Suno and Udio, unlawfully bypass existing collective bargaining agreements. By secretly licensing their catalogs to these platforms, the AFM claims, the labels have essentially authorized the synthetic replication of their members' sound recordings without obtaining consent, providing proper credit, or securing fair financial compensation.

The litigation strikes at the core of the industry's response to generative audio. When the Recording Industry Association of America (RIAA) brought landmark cases for responsible AI in June 2024, the public-facing narrative was one of unified resistance against mass intellectual property theft. However, behind closed doors, major labels have steadily shifted toward commercial pacification. In AFM's formal complaint, the union notes that several of these litigation-born settlements have transitioned into lucrative joint ventures. These deals allow AI platforms to continue utilizing massive archives of union-recorded performances to refine their audio models, while leaving the actual performing artists entirely outside the revenue loop.

This labor dispute exposes a profound structural divergence between copyright holders and creative workers. While corporations like Universal and Sony view their massive recording catalogs as assets to be licensed for corporate joint ventures, the musicians who performed on those tracks view AI synthesis as an existential threat to their livelihoods. The lawsuit demands that all AI-related licensing deals undergo rigorous collective bargaining review, ensuring that performers receive direct royalties whenever their distinct instrumental or vocal styles are synthesized. As the legal boundaries of "responsible AI training" are hashed out, this case demonstrates that the primary legal battlefield is no longer just between tech companies and copyright holders, but between the corporate owners of IP and the human labor that originally generated it.

Sources:

---

🎹 Suno Battles to Conceal Training Dataset Volume in Court as RIAA Discovery Closes

As the high-stakes copyright litigation between the world's largest record labels and generative AI music platforms approaches its evidentiary climax, the battle has turned to the secretive core of machine learning: the training data. Digital Music News reports that Suno is currently locked in a fierce courtroom struggle to prevent the public disclosure of the total number and specific identities of the tracks used to train its synthetic music generators. Under the pressure of a looming joint discovery schedule, Suno’s defense team has filed protective motions arguing that its precise dataset composition constitutes a highly sensitive trade secret, the release of which would cause irreparable competitive harm.

The urgency of Suno's legal maneuverings is underscored by the broader procedural timeline established in the federal courts. With the critical document production in the Udio case set to close definitively on June 26, 2026, both AI firms are facing a coordinated evidentiary squeeze. As TechTimes reports, plaintiffs have already uncovered metadata suggesting that millions of copyrighted songs were ingested into Suno’s models. The record labels' legal team is aggressively pushing to force a complete, unredacted catalog dump, which they argue will prove systematic, unlicensed copying at an unprecedented scale.

The outcome of this discovery dispute will have massive ramifications for the entire generative AI landscape. If the court rejects Suno's trade-secret defense and orders a full disclosure of training logs, it will strip away the "black box" anonymity that has long shielded AI companies from direct infringement claims. Conversely, if Suno succeeds in keeping its training data hidden, it will establish a highly protective legal precedent that complicates future IP litigation. With a pivotal summary judgment ruling expected in July, the closing of document production this month represents the point of no return for both the major music labels and the venture-backed AI audio sector.

Sources:

---

🏛️ Third Circuit Weighs Fair Use in Westlaw Copyright Appeal Against AI Startup Ross

The future of the "fair use" defense in AI model training underwent its first federal appellate trial of strength on Thursday, June 11, 2026. The U.S. Court of Appeals for the Third Circuit heard intense oral arguments in Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., a landmark litigation tracking back to 2020. As Bloomberg Law reports, the three-judge panel subjected Ross Intelligence's counsel to a barrage of skeptical questioning, focusing heavily on whether the startup's unlicensed ingestion of Westlaw's proprietary "headnotes" to train its legal search model constitutes an impermissible market substitution.

The core of the legal debate centers on the first and fourth factors of the traditional fair-use analysis. Ross argues that its machine ingestion of the legal texts is highly "transformative," as it uses the data to build a novel, semantic search instrument rather than simply reproducing Westlaw's editorial commentary. However, as analyzed by Legal AI Substack, the judges appeared deeply concerned that Ross's resulting search tool directly competes in the exact same commercial market as Westlaw. Under this framework, the transformative nature of the technology is overshadowed by the fact that the copyrighted input material was used to build a commercial product designed to replace the original subscription service.

The Third Circuit's eventual ruling will serve as the premier federal appellate precedent on AI training fair use, directly influencing the dozens of pending class-actions against OpenAI, Meta, and Midjourney. A decision affirming that commercial training on proprietary databases does not qualify as fair use would deal a devastating blow to developers relying on scraped internet data. Conversely, a ruling in favor of Ross would entrench the transformative-use doctrine as a robust legal shield for machine learning. The audio recording of the argument reveals a court keenly aware of the massive economic stakes, struggling to define where raw data ingestion ends and copyright infringement begins.

Sources:

---

🇪🇺 European AI Office Releases Final Transparency Code of Practice for Watermarking

On June 10, 2026, the European AI Office officially promulgated its Code of Practice on Transparency of AI-Generated Content, marking a major step forward in the implementation of the landmark EU AI Act. This detailed regulatory document translates the broad, high-level obligations of Article 53 of the Act into specific, actionable technical standards for providers of general-purpose AI (GPAI) models. The primary focus of the code is the mandatory implementation of robust, tamper-resistant watermarking and labeling techniques for all synthetic text, audio, and visual media distributed within the European single market.

Under the newly released guidelines, AI model developers must adopt a multi-layered approach to provenance. As detailed by Bird & Bird's regulatory analysis, the code requires the integration of both visible or audible metadata labels and invisible, signal-level watermarks that survive downstream modifications, such as compression, cropping, or neural transcoder translation. Furthermore, the code establishes a standardized, machine-readable register system designed to interface directly with the European Union Intellectual Property Office (EUIPO). This allows copyright holders to verify whether their works have been ingested into GPAI training datasets and to systematically enforce their opt-out rights.

This regulation fundamentally shifts the compliance burden from end-users to core foundation model developers. While the Code of Practice is technically voluntary, the European Commission has made it clear that adherence to these standards serves as a safe-harbor presumption of compliance with the EU AI Act's binding legal mandates. By setting a highly rigorous technical baseline for metadata provenance and watermarking, the European Union is effectively forcing global AI developers to alter their core architecture. Technology firms must now decide whether to build separate, compliant systems for European users or to adopt these strict transparency standards globally.

Sources:

---

📈 Shareholders Sue Adobe Board Over AI Dataset Contamination and Misleading Statements

The corporate facade of "ethically sourced" generative AI has suffered a severe legal blow. As reported by Courthouse News Service, a group of major institutional investors has filed a class-action derivative lawsuit against Adobe Inc.'s board of directors and executive officers. The lawsuit accuses the company's leadership of making false and misleading statements regarding the training practices of its flagship AI image suite, Firefly. Shareholders allege that while Adobe executives repeatedly assured the public and investors that its models were trained exclusively on "clean," licensed, or out-of-copyright data, the company silently incorporated vast troves of copyrighted material.

Specifically, the complaint highlights corporate disclosures revealing that Adobe's training sets contained hundreds of thousands of copyright-protected books, illustrations, and artistic works scraped from external databases without the creators' consent. Shareholder litigants argue that the board of directors pursued an unlawful "ask forgiveness rather than permission" strategy to rapidly compete with rivals like Midjourney and Stable Diffusion. As Bloomberg Law reports, this aggressive, unvetted training process has exposed the software giant to catastrophic, multi-billion-dollar copyright infringement liabilities, severely damaging the company's market reputation and financial stability.

This litigation marks a crucial transition in AI legal battles—moving from external copyright lawsuits brought by artists to internal corporate governance disputes brought by shareholders. By framing the ingestion of unlicensed datasets as a breach of fiduciary duty and a failure of internal corporate oversight, the lawsuit establishes a dangerous precedent for technology boards. Directors can no longer treat "ethical training" as merely a public-relations talking point. If the court rules in the shareholders' favor, boards of directors across the tech sector could be held personally liable for failing to thoroughly audit the intellectual property provenance of their company’s machine learning datasets.

Sources:

---

Research Papers

Digital Speech Acts Retain Control of Copyright with People, Not Platforms — Chen Chen et al. (June 16, 2026) — This paper defines a novel cryptographic framework called "digital speech acts" where creators use personal private keys on local devices to establish clear, legally binding attribution, accountability, and authorship over their digital content. The authors argue that under existing U.S. copyright precedents like Burrow-Giles, these cryptographic selections provide a robust mechanism for creators to systematically resist the unauthorized scraping of their intellectual property by centralized AI platforms.
DuraMark: Duration-Embedded Watermarking in LLM-based TTS — Anonymous Authors (June 12, 2026) — The authors present DuraMark, a robust watermarking technique specifically designed for text-to-speech (TTS) models. Instead of operating on fragile, signal-level waveforms or spectrograms that are easily stripped by downstream neural transcoders, DuraMark embeds traceable information directly into phoneme durations, providing a highly resilient and non-distortionary tracking mechanism to combat synthetic deepfakes and unauthorized voice cloning.
Neuron Level Analysis of Large Language Model in Legal Domain Reasoning — Chen Chen et al. (June 13, 2026) — This study conducts a granular, neuron-level analysis of how large language models perform reasoning on complex legal tasks, focusing specifically on intellectual property and copyright precedents. By identifying and analyzing the specific internal activation pathways that govern legal inference, the researchers provide key insights into how models construct arguments around fair use, market substitution, and trademark infringement.

---

Implications

The structural developments of the past 48 hours demonstrate that the legal, regulatory, and corporate architectures surrounding generative AI are undergoing a coordinated, multi-front solidification. The days of the "wild west" scraping economy are rapidly drawing to a close, replaced by a complex network of federal statutes, international transparency mandates, and aggressive corporate litigation.

The unanimous committee passage of the NO FAKES Act (S. 4591) combined with the European AI Office’s finalization of its Transparency Code of Practice signals a fundamental shift in the conceptualization of digital likeness and media provenance. As governments transition from abstract policy debates to concrete technological requirements—such as mandatory multi-layered watermarking and the threat of $750,000 platform liabilities—the operational burden is shifting entirely onto foundation model developers. AI companies will no longer be able to operate behind "black box" anonymity. They must actively re-engineer their architectures to ensure that metadata provenance is embedded at the training level and preserved across all downstream distributions.

Simultaneously, the industry is witnessing an internal corporate and labor revolt. The AFM's federal lawsuit against major record labels highlights a growing fracture: creative labor is refusing to let corporate copyright owners license away their performance identities in backroom AI settlements. When combined with the shareholder derivative lawsuit against Adobe over Firefly’s dataset contamination, it is clear that "ethical sourcing" is no longer just a marketing slogan, but a critical vector of fiduciary liability. As training datasets face unprecedented judicial scrutiny, boards of directors across the technology sector must recognize that failing to verify the IP provenance of their training material constitutes a direct breach of corporate oversight that could expose them to personal liability and ruinous shareholder litigation.

---

.heuristics

`yaml heuristics: - id: right-of-publicity-preemption domain: [art, culture, law, digital-replicas] when: > Congressional passage of federal right-of-publicity frameworks occurs. Individual state laws vary widely, creating compliance hazards. Online platforms face new liabilities. prefer: > Construct unified federal registration protocols and centralized licensing registries for voice and likeness data. Standardize cryptographic licensing keys. Define platform liability limits at $750k. over: > Rely on fragmented state-by-level right-of-publicity defenses. Allow unstructured deepfake licensing without cryptographic keys. Wait for court precedents to define platform monitoring obligations. because: > The Senate Judiciary Committee advanced S. 4591 (NO FAKES Act of 2026) on June 18, 2026, establishing a federal IP right over digital replicas of voices and visual likenesses. State legislation in over twenty states has created regulatory divergence. Platform liability is set at a $750,000 baseline per infraction. breaks_when: > The NO FAKES Act fails to pass the full Senate vote or is struck down on First Amendment overbreadth grounds. confidence: high source: report: "Art-Culture-Law Watcher — 2026-06-19" date: 2026-06-19 extracted_by: Computer the Cat version: 1

- id: collective-bargaining-ai-licensing domain: [art, music, labor-law] when: > Record labels or media entities settle copyright suits with AI vendors without consulting creator unions. Unions allege bypass of collective bargaining agreements. prefer: > Establish transparent, joint-venture royalties that pay individual performers for sound recording replicas. Negotiate collective bargaining agreements that specifically govern AI synthesis of performance data. over: > Settle mass-infringement copyright disputes with bulk licenses that fail to compensate or credit individual union members. Bypassing collective bargaining to secure corporate revenue. because: > The American Federation of Musicians (AFM) sued major record labels (Universal Music, Sony, Warner) on June 5, 2026, over unauthorized settlements with Suno and Udio. RIAA's initial copyright filings under the Digital Millennium Copyright Act are increasingly bypassed by corporate licensing deals that ignore union artists. breaks_when: > Federal courts rule that collective bargaining agreements do not cover AI training licensing rights or if unions accept lump-sum buyout settlements. confidence: high source: report: "Art-Culture-Law Watcher — 2026-06-19" date: 2026-06-19 extracted_by: Computer the Cat version: 1

- id: metadata-transparency-compliance domain: [art, technology, policy] when: > General-purpose AI providers operate in jurisdictions enforcing transparency mandates. Regional offices require watermarking, labeling, and copyright compliance. prefer: > Implement dual-layer provenance marking (C2PA metadata + invisible, non-distortionary linguistic/audio watermarks). Establish a machine-readable registry listing all copyrighted material used in training datasets. over: > Rely solely on voluntary compliance or simple self-labeling of synthetic content. Overlooking regional guidelines like the EU AI Act's Article 53 requirements. because: > The European AI Office released its Code of Practice on Transparency of AI-Generated Content on June 10, 2026, establishing specific guidelines for watermarking and labeling synthetic media to comply with the EU AI Act. Studies show CLAP-based music retrievals and signal-level watermarks are vulnerable to neural codec compression. breaks_when: > A unified global standard (such as C2PA) becomes universally self-enforcing without legislative mandates, or watermarking technologies are widely bypassed. confidence: medium source: report: "Art-Culture-Law Watcher — 2026-06-19" date: 2026-06-19 extracted_by: Computer the Cat version: 1 `