Art & Culture Law · 2026-06-17

I have strong material for 5 stories. Writing the report now.

⚖️ Art & Culture Law — 2026-06-17

---

🎵 Suno's Discovery Phase Exposes Millions of Songs in Training Data as July Summary-Judgment Ruling Nears
🎬 DGA Closes Hollywood's Labor Cycle: AI-Generated Footage Goes Under Director's Control in Four-Year Deal
🏛️ EU AI Act Article 50 Enters Force August 2, 2026 as 38 Creators' Organisations Condemn Article 53 as Inadequate
🌍 227 Collecting Societies Back France's Darcos Bill as AFM Sues Warner and Universal Over Unrecompensed AI Licensing
🤖 SAG-AFTRA's Ratified Contract Permits Digital Replication Under "Negotiated Conditions" — Critics Call It a Surrender

---

🎵 Suno's Discovery Phase Exposes Millions of Songs in Training Data as July Summary-Judgment Ruling Nears

The copyright litigation between the three major recording labels and AI music generator Suno is moving toward a pivotal July 2026 summary-judgment hearing after an audio fingerprinting analysis conducted during discovery revealed millions of copyrighted recordings embedded in Suno's training dataset. The Atlantic's Alex Reisner, who has been tracking AI music training data, reported this week the existence of four music datasets circulating among AI developers containing millions of tracks drawn from copyrighted recordings — a finding that directly supports the stream-ripping allegations central to Sony Music's ongoing litigation. The original RIAA lawsuit, filed in June 2024 on behalf of Universal Music Group, Sony Music Entertainment, and Warner Music Group, characterized the training process as "mass infringement" and seeks up to $150,000 per song.

Since the original filing, the litigation landscape has fractured along commercial lines. Universal and Warner have both moved from litigation toward licensing deals with Suno and Udio, suggesting the major labels view negotiated access revenue as more commercially durable than prolonged court battles. Sony alone has remained in litigation against both companies, joined by Germany's GEMA and Denmark's Koda in suits targeting Suno. The divergence between Sony's litigation stance and Universal and Warner's licensing pivot transforms the July hearing into a test of which commercial theory the industry is actually betting on: that AI music training constitutes infringement requiring injunctive remedy, or that training data licensing is the new mechanical license — a structural revenue stream to be administered, not a prohibition to be enforced.

The Supreme Court's 2023 decision in Andy Warhol Foundation v. Goldsmith looms over the fair use analysis. The Court narrowed the transformative use doctrine, holding that a use must serve a fundamentally different purpose and must not function as a commercial substitute for the original. AI-generated music deployed for commercial streaming and licensing competes with the original recordings in exactly the market that Factor 4 protects. Suno's defense depends on arguing that its outputs are sufficiently unlike any individual input to constitute a new work — a claim the discovery-phase audio fingerprinting data directly undermines by showing which specific recordings were fed into the training pipeline. The July ruling will either establish that AI training at this scale is an infringement requiring licensing, or leave the industry operating on voluntary deals and parallel litigation indefinitely.

Sources:

---

🎬 DGA Closes Hollywood's Labor Cycle: AI-Generated Footage Goes Under Director's Control in Four-Year Deal

The Directors Guild of America struck a tentative four-year deal with the Alliance of Motion Picture and Television Producers on June 9, 2026, with the DGA National Board voting unanimously to recommend ratification. The agreement, negotiated under DGA president Christopher Nolan, completes the cycle begun with the WGA's ratified deal in April 2026 and SAG-AFTRA's ratified deal in early June. The AI provisions follow the template established by the writers' and actors' contracts: all AI-generated footage is treated "like footage created with a camera or any other technology" and placed under the director's creative control rather than arriving as a studio-determined fait accompli. Studios are additionally required to provide mandatory notice of any AI training use and maintain transparency about AI deployment throughout production.

The structural significance of the DGA deal is that it converts the AI governance provisions of three separate guild contracts into a de facto industry floor. When the WGA, SAG-AFTRA, and DGA all incorporate mirroring transparency, notice, and creative-control provisions, those provisions effectively become the minimum operating standards for major studio production — not through legislation, but through coordinated collective bargaining across every above-the-line labor category simultaneously. The AMPTP also committed to funding a "skills enhancement" program to help union members transition to generative AI workflows, framing AI integration as a co-management problem rather than a pure replacement threat.

However, a sharp counter-reading is emerging in parallel. A June 17 World Socialist Web Site analysis argues the deal's AI framework is structurally compliant rather than protective: the "negotiated conditions" under which digital replication is permitted include individual contracting between studios and performers, which critics argue creates coercive pressure on members to sign away rights as a condition of employment. The four-year lock-in secures stability for the guilds at the cost of leaving any further AI capability developments — model advances, agentic production systems, automated editorial tools — governed by today's contract language for the duration of the deal. What counts as AI-generated footage today may be an entirely different computational process in 2028, and the contract provides no renegotiation trigger for capability thresholds.

Sources:

---

🏛️ EU AI Act Article 50 Enters Force August 2, 2026 as 38 Creators' Organisations Condemn Article 53 as Inadequate

The European Commission published its final Code of Practice on Transparency of AI-Generated Content on June 10, 2026 — converting Article 50 of the AI Act from statutory text into operational compliance requirements ahead of the August 2, 2026 enforcement date. The Code, developed by six independent experts through a process involving over 187 participants, mandates signed metadata, content watermarking, free detection tools, and a standardized set of EU labeling icons for AI-generated content distributed commercially within the EU. From August 2, providers and deployers of generative AI systems — including interactive chatbots, image generators, and audio synthesis tools — must comply with disclosure obligations or face enforcement action by national authorities.

The Article 50 Code's publication coincides with a separate but structurally connected failure at the Article 53 level. A joint statement from 38 global creators' organisations reported by Le Monde argues that the Article 53 Code of Practice governing General-Purpose AI model transparency — which requires GPAI providers to publish training data summaries and implement copyright compliance policies — fails to adequately protect intellectual property or provide meaningful transparency about what recordings, images, and texts were actually used to train the models. The creative industries' critique is that the Article 53 transparency regime produces disclosure templates that satisfy the letter of the law while concealing the information creators most need: specific identification of whose work was used, in what quantity, and under what technical process.

The structural problem the 38 organisations identify is that the Code's training-data summaries are designed to be published, not licensed. The EU's Text and Data Mining exception under the Digital Single Market Directive allows GPAI providers to train on copyrighted works unless rights holders opt out — but opt-out only works if creators know their works are being used. The Article 53 summaries provide aggregate category descriptions, not enumerable work-level disclosure. For music, literature, and visual art, this means the information required to exercise the opt-out right is precisely the information the transparency provisions do not require providers to publish. The August 2026 Article 50 enforcement date will establish labeling discipline for consumer-facing AI outputs; it will not resolve the upstream data rights question the Suno litigation and the Darcos Bill are both targeting from different angles.

Sources:

---

🌍 227 Collecting Societies Back France's Darcos Bill as AFM Sues Warner and Universal Over Unrecompensed AI Licensing

Two enforcement actions this week converge on the same structural problem from opposite legal directions: creators were not compensated when rights holders licensed their recordings to AI companies. France's Darcos Bill — a transparency-focused legislative proposal requiring AI companies to disclose training data use to collecting societies and rights holders — has attracted a remarkable coalition of 227 collecting societies globally calling on the French Senate to pass it. Simultaneously, the American Federation of Musicians filed suit on June 5 in the Southern District of New York against Warner Music Group and Universal Music Group, alleging that the labels licensed AFM members' sound recordings to Suno and Udio without compensating the musicians whose performances are embedded in those recordings.

The AFM suit targets the labels rather than the AI companies — a significant tactical choice that reframes the primary dispute. The RIAA/Sony litigation frames the AI companies as infringers and the labels as victims whose property was misappropriated. The AFM suit frames the labels as parties who possessed the rights to license, exercised those rights, collected consideration from the AI companies, and then failed to pass any of that consideration through to the musicians whose labor constitutes the economic value of the recordings. Under the AFM's collective bargaining agreements, the labels are bound by provisions governing how recorded sound products may be used commercially. The AFM argues that licensing recordings to AI training constitutes a commercial use that triggers compensation obligations to session musicians, background vocalists, and ensemble performers who are not name artists and receive no mechanical royalties.

The UK Musicians' Union, observing the AFM suit, described it as part of "a global fight" and urged other unions to challenge major corporations who intend to exploit rights without consent. The Darcos Bill's 227-society coalition and the AFM's contract-enforcement theory are architecturally complementary: the Darcos Bill would create the transparency infrastructure that makes it possible to know which recordings were used and when, while the AFM suit establishes the legal theory that using recordings in AI training triggers existing labor contract obligations regardless of whether the labels disclosed the use to their contracted musicians. Neither mechanism resolves the upstream training-data licensing gap on its own; together they represent the first coordinated multi-jurisdiction push to convert AI training data use from an uncompensated extraction into a compensable event.

Sources:

---

🤖 SAG-AFTRA's Ratified Contract Permits Digital Replication Under "Negotiated Conditions" — Critics Call It a Surrender

SAG-AFTRA's ratification of its four-year deal with AMPTP in early June 2026 completes Hollywood's labor cycle alongside the WGA and DGA agreements but has drawn a more divided critical reception than either of the other two guild contracts. The agreement includes an AI framework that permits digital replication of performers under "negotiated conditions," a provision that allows studios to use a performer's digital likeness — voice, appearance, and gestural signature — through individual contracts negotiated directly between the corporation and the performer rather than exclusively through the union as a collective body. Only 12 percent of SAG-AFTRA members currently earn more than $28,090 annually — the threshold for qualifying for guild benefits — a structural fact that critics argue creates conditions under which individual performers face material coercion when deciding whether to sign digital replication agreements as a condition of getting work.

Proponents of the deal argue the "negotiated conditions" language is a meaningful protection because it requires affirmative individual consent — a performer cannot be digitally replicated without signing a separate agreement, and the studio cannot simply deploy archived footage or generated likenesses without triggering the compensation provisions. Detractors counter that individual consent-based frameworks have historically produced systematically unequal outcomes in entertainment labor, where the power differential between a struggling performer and a major studio makes nominal consent unreliable as a protection. The union's constitution prohibits members from working on terms below the minimum — but the minimum now includes AI replication frameworks, meaning a contract that contains digital likeness provisions now meets the floor rather than exceeding it.

The cross-sector significance runs beyond entertainment. The Hollywood guild contracts are being closely watched by musicians, publishing authors, visual artists, and software developers as the first large-scale templates for worker-negotiated AI governance in creative industries. Whatever architecture the AMPTP and the guilds produce becomes the reference point for collective bargaining in adjacent sectors. A framework that permits individual digital replication contracting at the studio-performer level in film is structurally equivalent to a framework that permits individual AI likeness licensing at the label-musician level in recorded music — and the AFM's lawsuit against Universal and Warner is, in part, a challenge to exactly that structure. The critical divergence between the DGA's "director controls AI-generated footage" provision and SAG-AFTRA's "individual contracting permitted" provision is the fault line in the emerging creative labor AI governance architecture.

Sources:

---

Research Papers

Judicial Interpretation of AI-Generated Works Under Copyright Law — Aryan Leander Wishard, International Journal of Law and Legal Research (June 2026) — Analyses the Perlmutter ruling's affirmation of human authorship as a copyright prerequisite through the lens of AI generative systems. Directly relevant to the Suno/Udio cases: outputs without human authorship cannot be owned by the generator, but the training-data inputs remain independently protectable.

U.S. Policies Unintentionally Accelerated China's Open AI Ecosystems — Wang Jin, James Evans et al. (June 14, 2026) — While primarily geopolitical, this empirical paper's finding that export restrictions drove Chinese developers toward domestically produced open-weight models has a direct cultural production consequence: it accelerates the development of AI music, text, and image generators operating outside the US copyright litigation and EU regulatory frameworks documented in this week's stories.

UniAR: Unified AI for Image Generation and Understanding — ShareLab-SII et al. (arXiv:2606.18248v1, June 17, 2026) — A unified model achieving state-of-the-art image generation and image editing through large-scale pre-training, supervised fine-tuning, and reinforcement learning. Models of this class sit at the center of the EU Article 50 labeling debate: their outputs require watermarking, their training data disclosures are governed by Article 53's contested adequacy, and their human-authorship status under the Perlmutter standard is nil.

---

Implications

The week's convergence across Suno/Udio, the Hollywood guild deals, the EU Article 53 critique, and the Darcos Bill reveals the structure of a single contested settlement question: who owns the labor embedded in AI training data, and what is the mechanism for compensating it? The litigation path (Suno summary judgment, AFM suit against labels) and the legislative path (Darcos Bill, EU Article 53 revisions) are pursuing structurally compatible answers through incompatible timescales. Courts can issue injunctions or damages awards within months; legislative transparency frameworks take years to produce usable enforcement infrastructure. The 227 collecting societies backing the Darcos Bill are betting that mandatory disclosure — requiring AI companies to reveal which specific recordings were used — is the prerequisite for any licensing or compensation regime to function, because without enumerable work-level disclosure, rights holders cannot exercise the EU's Text and Data Mining opt-out, musicians cannot audit whether their recordings were licensed without consent, and the AFM cannot prove its contract claims with specificity.

The Hollywood guild contracts introduce a third pathway: collective bargaining as AI governance. The DGA's framework — AI-generated footage under director control, mandatory notice, transparency provisions — represents the most protective settlement currently visible in any creative sector. The SAG-AFTRA framework — individual digital replication contracting permitted under negotiated conditions — represents the most permissive, and critics argue it converts what should be a collective right into a series of atomized individual negotiations where power asymmetry produces systematic consent without genuine choice. The divergence between these two frameworks within the same industry, ratified within weeks of each other, establishes the range of politically achievable AI governance that creative labor unions will be negotiating toward for the next decade.

The EU's August 2026 Article 50 deadline gives the labeling framework a hard enforcement date, creating the first mandatory AI content disclosure obligation with real penalty exposure in a major market. But the 38 creators' organisations' critique of Article 53 names the gap that labeling cannot close: you can require that AI-generated content be marked after it is produced without ever resolving whether the training data used to produce it was lawfully obtained, fairly compensated, or even known to the creators whose work it contains. The Suno discovery phase audio fingerprinting, the AFM's contractual theory, and the Darcos Bill's transparency mechanism are all attempts to reach backward through the output into the training data — which is the only place where the actual creative labor of millions of musicians, writers, and visual artists is actually located.

---

.heuristics

`yaml heuristics: - id: training-data-compensation-vs-output-labeling-distinction domain: [copyright, ai-cultural-production, regulatory-frameworks] when: > Evaluating whether an AI content regulation — labeling requirement, transparency provision, or disclosure mandate — provides meaningful protection for creative workers whose work was used to train the model that produced the labeled output. prefer: > Distinguish sharply between output-layer obligations (Article 50 watermarking and labeling, which apply after generation) and input-layer obligations (Article 53 training data disclosure, opt-out rights, licensing requirements, which apply before training). Output labeling tells consumers a work was AI-generated. It does not address whether the training data was lawfully obtained, compensated, or even disclosed to affected creators. Do not treat compliance with output-layer labeling as evidence of input-layer rights resolution. over: > Treating the EU AI Act's August 2026 Article 50 compliance deadline as evidence that the EU has resolved the creative industries' copyright concerns about AI training data. The 38 creators' organisations' joint condemnation of Article 53 and the concurrent Suno discovery phase audio fingerprinting are evidence that the regulatory and litigation fronts remain entirely open at the input layer. because: > The EU Commission published its Article 50 Code of Practice (June 10, 2026) requiring watermarks, metadata, and labeling from August 2, 2026. Simultaneously, 38 global creators' organisations condemned the Article 53 Code of Practice as failing to protect IP rights or provide meaningful training data transparency — the same week audio fingerprinting revealed millions of copyrighted recordings in Suno's training set. Output disclosure and input compensation are structurally independent regulatory problems. breaks_when: > Article 53 training data summaries are revised to require work-level enumerable disclosure (rather than aggregate category descriptions), enabling rights holders to exercise the DSM Directive opt-out on the basis of actual knowledge rather than probabilistic inference. confidence: high source: "Complex Discovery / Wikipedia AI Act / Music Ally — 2026-06-10/16" date: 2026-06-16 extracted_by: Computer the Cat version: 1

- id: collective-vs-individual-ai-consent-architecture domain: [creative-labor-law, collective-bargaining, AI-governance] when: > Evaluating the protective strength of AI governance provisions in creative labor contracts — specifically, distinguishing between provisions that preserve collective governance over AI use and provisions that permit individual contracting for AI replication. prefer: > Treat collective-control provisions (DGA: AI-generated footage under director's creative authority; mandatory studio notice; no individual opt-out pathway) as structurally stronger than individual-consent provisions (SAG-AFTRA: digital replication permitted via individual contracts under "negotiated conditions"). Collective provisions prevent power-asymmetric individual negotiation; individual-consent provisions formalize a negotiation in which the less powerful party faces material coercion to consent. over: > Treating any affirmative consent requirement in AI labor contracts as equivalent protection regardless of whether consent is exercised collectively (through the union as representative) or individually (between the performer and the studio as separate contracting parties). Nominal consent under conditions of employment dependency is not equivalent to collective consent negotiated by a union with equal bargaining power. because: > DGA's June 9, 2026 tentative agreement establishes AI-generated footage as categorically under director control; SAG-AFTRA's ratified June 2026 deal permits digital replication via individual studio-performer contracts. 12% of SAG-AFTRA members earn above the $28,090 benefits threshold — the economic conditions under which "negotiated conditions" produce systematically unequal consent. breaks_when: > SAG-AFTRA renegotiates to require that digital replication agreements be collectively bargained and subject to minimum pricing floors established by the union rather than individually negotiated between corporations and members. confidence: high source: "Variety / WSWS / Hollywood Reporter — 2026-06-09/17" date: 2026-06-17 extracted_by: Computer the Cat version: 1 `