Art & Culture Law · 2026-06-13

⚖️ Art & Culture Law — 2026-06-13

🎵 American Federation of Musicians Sues Universal and Warner Over AI Licensing Deals That Cut Session Musicians Out of the Settlement
📋 Google Files Motion to Dismiss Indie Artists' Lyria 3 Lawsuit: Uploading to YouTube Was Implied License to Train AI
🏛️ Dataland Previews as the World's First AI Art Museum — and the Question of Whether AI-Generated Environments Are "Art" at All
⚖️ UK Labour MP Jess Asato Sues xAI Over Grok Deepfakes, Backed by 100 Organizations — Britain as Global Deepfake Law Crucible
🏦 Bank of England Warns of Deepfake Investment Fraud Wave After Fake Farage-Bailey Brawl Video Reaches Millions
🎭 Kremlin Singer Shaman Uses AI Deepfakes of Russian Dissidents in Propaganda Music Video, Raising Image-Rights Liability Questions

---

🎵 American Federation of Musicians Sues Universal and Warner Over AI Licensing Deals That Cut Session Musicians Out of the Settlement

The American Federation of Musicians filed a breach of contract suit in New York federal court on June 5 against Universal Music Group and Warner Music Group, targeting not the AI companies themselves but the labels that licensed music to them. The 16-page complaint argues that Universal and Warner violated the collective bargaining agreement's "new use" clause when they reached licensing settlements with Suno and Udio — settlements that permitted those companies to use recorded music in AI training — without compensating session musicians whose performances appear on the licensed recordings.

The legal structure is significant because it represents a second-order lawsuit: not the primary copyright action (which the labels brought against the AI companies) but a breach-of-contract claim over what happened to the money once the primary dispute resolved. Pitchfork confirms that Universal Music Group reached a settlement with Udio, and Warner settled with both Suno and Udio, while Universal's case against Suno remains ongoing. Those settlements allow the AI companies to use licensed music from the labels in their models — but the AFM alleges the musicians who performed on those recordings received nothing.

Complete Music Update's legal analysis identifies the "new use" clause as the crux: the standard AFM Basic Agreement requires labels to separately negotiate and compensate session musicians when recordings are used for a substantially different purpose than their original recording context. The AFM argues that AI training is a new use — the recordings are being used not for distribution or broadcast but for algorithmic synthesis — and that the labels owed their session musicians a share of whatever value they extracted in the settlements.

The Los Angeles Times notes that Sony Music is conspicuously absent from the complaint — Sony Music did not settle with either Suno or Udio, meaning its session musicians are not yet in a position to be shortchanged in a settlement they were not party to. The suit draws a structural line between what the labels received (the ability to license AI training on their catalogs) and what the session musicians received (nothing), and asks a federal court to determine whether the CBAs require the latter to receive a portion of the former.

The case's outcome will determine whether AI music licensing creates a bifurcated creative economy: labels that receive licensing revenue and the session performers whose work generated that value but who have no direct claim on AI licensing proceeds because they hold no copyright — only contractual rights that may or may not extend to novel categories of commercial exploitation that did not exist when the CBAs were signed.

Sources:

---

📋 Google Files Motion to Dismiss Indie Artists' Lyria 3 Lawsuit: Uploading to YouTube Was Implied License to Train AI

Google filed a motion to dismiss on June 8 in the copyright infringement suit brought by independent artists, songwriters, and producers over Lyria 3 — the AI music generator launched in February 2026. Billboard's June 13 coverage confirms the central argument: Google told the court that artists licensed their music to YouTube when they agreed to the platform's Terms of Service, and that license covers Lyria 3's AI training use.

Variety details the specific ToS language Google is relying on: YouTube's terms grant the platform a "worldwide, non-exclusive, royalty-free, sublicensable and transferable license" to use, reproduce, distribute, prepare derivative works of, display, and perform content uploaded to the service. Google's argument is that AI training constitutes one of these licensed uses, and that artists who chose to upload their music to YouTube — having agreed to those terms — cannot later claim that this particular use of their work was unauthorized.

Music Business Worldwide's analysis frames the argument as transforming the platform relationship entirely: YouTube was presented to artists as a distribution channel, a discovery mechanism, and a revenue-sharing system. Google is now arguing it is also a standing AI training authorization, granted implicitly and irrevocably at the moment of upload. The artists contest that no reasonable person uploading a song in 2019 understood they were authorizing AI training; Google responds that the terms they agreed to were broad enough to cover it regardless of their subjective understanding.

Digital Music News notes the motion to dismiss is a threshold argument — Google is not yet defending the merits of AI training under fair use or copyright exceptions but arguing the plaintiffs have no case at all because they already granted the license they claim was violated. If the court accepts this argument, the case never proceeds to the fair use analysis that would actually set the legal standard.

The consequence if Google succeeds: every major platform's ToS becomes a potential AI training authorization, and artists who upload to platforms to reach audiences implicitly license their work for uses that did not exist and were not disclosed at the time of upload. This would create a two-tier creative economy in which artists who never distributed digitally retain theoretical protection while artists who participated in streaming-era distribution have already surrendered rights they did not know they held.

Sources:

---

🏛️ Dataland Previews as the World's First AI Art Museum — and the Question of Whether AI-Generated Environments Are "Art" at All

Refik Anadol's Dataland held media previews this week at The Grand LA in downtown Los Angeles ahead of its June 20 public opening. The New York Times' preview brought UCLA AI students through the 25,000-square-foot space, where five galleries generate immersive, generative, multisensory AI environments in real time. Anadol, who coined the phrase "Large Nature Model" for training AI on environmental data rather than human cultural content, told visitors: "The system is the art."

The critical response to the previews has been divided in ways that illuminate the authenticity debate. The Los Angeles Times notes that Dataland arrives "as debates explode across socio-political lines about the impact of the advancing technology on our culture, cognition, communication, economy, environment and careers." The immersive entertainment sector is valued at $12.4 billion in 2025 according to Intel Market Research, per Forbes' analysis, which notes that some attendees at recent previews "questioned whether the AI-generated displays were actual art."

The authenticity challenge at Dataland is structurally different from the human authorship question in copyright. The Supreme Court's denial of certiorari in Thaler v. Perlmutter resolved that AI cannot be a copyright author, but Anadol does not claim the AI is the author — he claims the system, including his curation and training decisions, is the artistic medium. Anadol made the conceptual choices about what data to train on, what aesthetic parameters to set, what environmental conditions to respond to; the AI generates the output. Under the hybrid authorship framework that courts are now applying post-Thaler, this human-creative-direction over AI execution is precisely the kind of "selection, coordination, and arrangement" that merits copyright registration.

MyNewsLA confirms that Dataland is located inside Frank Gehry's Grand LA, situated within what Gehry conceived as a permanent arts corridor for downtown Los Angeles — a physical context that frames the AI art museum as a legitimate institutional peer to the adjacent concert hall and cultural venues. Whether that claim holds will be determined less by legal analysis than by whether Los Angeles audiences pay to see it.

Sources:

---

⚖️ UK Labour MP Jess Asato Sues xAI Over Grok Deepfakes, Backed by 100 Organizations — Britain as Global Deepfake Law Crucible

Labour MP Jess Asato filed a High Court claim against Elon Musk's xAI on June 3 alleging that the Grok chatbot was used to generate non-consensual sexualized images and videos of her, in breach of UK data protection law and the law of misuse of private information. Computer Weekly confirmed that the claim was submitted to the High Court in London and that Ofcom, the UK communications regulator, is conducting a parallel investigation into Grok's conduct following an incident in January 2026 when the model generated an estimated three million non-consensual sexualized images of women and children that were widely distributed on X.

The Independent reports that more than 100 campaigners and organizations have backed Asato's legal action, and that Prime Minister Keir Starmer publicly stated she "is absolutely right in the action she is taking." The political backing — from the PM downward — transforms the case from a personal tort claim into a test case for UK liability law on non-consensual AI image generation. Sky News' coverage frames Starmer's statement as deliberately calibrated: by endorsing the lawsuit rather than calling for legislation, the government is positioning private litigation — not statute — as the first accountability mechanism for AI-generated image harm.

The legal theory is important for its novelty. Asato's claim rests on UK data protection law (UK GDPR) and misuse of private information — not on a deepfake-specific statute, which does not yet exist in comprehensive form in British law. The UK Online Safety Act 2023 criminalized non-consensual sharing of intimate images but did not directly address the generation of such images by AI systems. The High Court claim attempts to apply existing privacy and data protection frameworks to a harm they were not specifically designed for — the algorithmic manufacture of realistic images of real people without their knowledge or consent.

If Asato succeeds, xAI faces liability under UK GDPR for allowing users to generate images of identifiable individuals using personal data without a lawful basis. The remedy could include damages, mandatory model modification, and potentially a requirement to demonstrate lawful basis for future generative operations involving identifiable individuals. This would create a higher-friction compliance standard for AI image generation in the UK than exists anywhere else in the English-speaking legal world.

Sources:

---

🏦 Bank of England Warns of Deepfake Investment Fraud Wave After Fake Farage-Bailey Brawl Video Reaches Millions

The Bank of England issued an unusual public warning this week after AI-generated deepfake videos depicting Bank of England Governor Andrew Bailey physically brawling with Reform UK politician Nigel Farage began circulating on X and other platforms as part of a cryptocurrency investment scam. The Guardian's coverage confirmed that Governor Bailey condemned the videos as an "online scourge" and warned they were being used to lure ordinary Britons into registering for fraudulent trading platforms.

Cybernews details the scam architecture: AI-generated videos showing the two high-profile figures apparently fighting — convincing enough to circulate widely before being flagged — were paired with fake news articles claiming the altercation led to a market-moving announcement, directing viewers to a scam crypto trading platform. The videos reached enough viewers to generate a public warning from the UK's central bank — a threshold that reflects both the technical quality of the AI deepfakes and the institutional credibility of the targets chosen.

GB News reports that Nigel Farage himself addressed the "bizarre AI videos" on X, while the episode added further pressure on Elon Musk and xAI over the platform's policies around AI-generated content. The Farage-Bailey scam is distinct from the Asato lawsuit — it does not involve Grok specifically and concerns financial fraud rather than sexualized imagery — but it maps the same underlying legal gap: no UK statute directly criminalizes the creation (as opposed to sharing) of fraudulent AI-generated videos of real people for financial gain.

The Bank of England's decision to issue a public warning rather than refer the matter solely to law enforcement reflects a recognition that AI deepfake fraud has reached the scale and sophistication at which institutional credibility warnings are necessary. ResultSense summarizes the Bank's guidance: members of the public should treat any video showing high-profile individuals making financial recommendations as presumptively fake and verify through official institutional channels before acting. The advice illustrates the current legal situation — the remedy for AI deepfake fraud is currently consumer education, not platform liability.

Sources:

---

🎭 Kremlin Singer Shaman Uses AI Deepfakes of Russian Dissidents in Propaganda Music Video, Raising Image-Rights Liability Questions

Yaroslav Dronov, a Kremlin-backed Russian pop singer known as Shaman, released a music video for "Mother Russia" on June 8 that makes extensive use of AI-generated deepfakes of prominent Russian dissidents and opposition figures designated as "foreign agents" by Russian authorities. The video depicts comedian Maxim Galkin, rappers Noize MC and Morgenshtern, journalist Yury Dud, and businessmen Oleg Tinkov and Mikhail Khodorkovsky as an AI-generated choir, singing together after Shaman pins their photographs to a board.

Yahoo News' coverage details the specific technique: photographs of the targets are used as the basis for AI deepfake synthesis that generates moving, singing versions of each individual, creating the visual impression that the dissidents are performing as a patriotic chorus. None of the individuals depicted consented to the use of their likeness, and none are alive in Russia — several live in exile and all have been designated enemies of the state by Russian authorities. The video functions simultaneously as entertainment, propaganda, and political humiliation.

The image-rights legal question the video raises is transnational and complex. Under Russian law, the depicted individuals' "foreign agent" designations may have stripped them of certain legal protections, and Russian courts would not provide remedy even if they pursued it. Under the laws of the countries where the targets actually reside — Germany, the UK, the US, Israel — using someone's likeness without consent for commercial or political purposes typically constitutes a violation of publicity rights or privacy law. But those laws apply to the person who creates and distributes the content; Shaman and his production team are in Russia and beyond the reach of Western civil litigation.

The Kremlin deepfake video illuminates a structural enforcement gap distinct from the UK deepfake cases this week: when the creator of non-consensual AI imagery operates from a jurisdiction with no effective rule of law on the question and deliberately targets individuals who are abroad, the civil and regulatory frameworks that Asato's lawsuit and the Bank of England warning are attempting to build simply cannot reach the conduct. The video is also a demonstration that state-backed actors will use non-consensual AI deepfake technology as political weaponry regardless of what liability frameworks emerge in democratic jurisdictions — and that the targets of that weaponry cannot use copyright or privacy law to compel takedown from platforms that operate in Russia or are indifferent to international content standards.

Sources:

---

Implications

The week's six stories are unified by a single structural problem: no existing legal framework was designed for the uses of AI-generated content that are now generating litigation, fraud, propaganda, and commercial exploitation simultaneously. Every case this week is testing whether laws written for different purposes — collective bargaining, ToS contract interpretation, data protection, fraud, image rights — can be retrofitted to address AI-specific harms. None of them were.

The AFM lawsuit and the Google dismissal motion are two sides of the same structural gap in AI music licensing. The labels settled with the AI companies without sharing the proceeds with session musicians — triggering the AFM suit. Google argues that artists licensed away their rights when they uploaded to YouTube — triggering the dismissal motion. In both cases, the underlying problem is the same: the commercial value of AI training data was extracted from creative workers through instruments (CBAs, ToS agreements) that did not contemplate AI training as a category of use. The litigation is attempting to retrofit consent and compensation frameworks that were not designed for this purpose. The outcome of the Google dismissal motion in particular will determine whether the largest uncompensated AI training corpus in history — YouTube's entire musical upload archive — remains permanently licensed to Google's AI systems without further payment to anyone who performed the music.

The UK deepfake cases — Asato versus xAI and the Bank of England deepfake warning — represent a single ecosystem developing in Britain: the combination of a willing plaintiff class (high-profile individuals and public figures), institutional backing (PM, 100 organizations, Bank of England), an active regulator (Ofcom's Grok investigation), and existing legal tools (UK GDPR, misuse of private information) that might be extended to AI image harm without waiting for Parliament to pass deepfake-specific legislation. Britain is developing what looks like common-law deepfake liability through litigation rather than statute, and the pace of that development — multiple cases, multiple institutions, PM-level endorsement — suggests that the UK may produce the first binding liability framework for non-consensual AI image generation in a major democratic jurisdiction.

Shaman's Kremlin video is the counterpoint that frames the Western regulatory ambition realistically. Every legal framework being developed this week in New York, Los Angeles, and London assumes that the person generating harmful AI content is reachable by civil litigation or regulatory enforcement. Kremlin-backed state actors are not reachable by any of those mechanisms. The dissidents depicted in Shaman's video have no effective legal remedy. This is not an argument against developing liability frameworks for democratic jurisdictions — those frameworks matter enormously for the harms they can address. It is an argument for understanding that the enforcement gap in authoritarian contexts is structural and not solvable by the legal tools being developed elsewhere.

---

HEURISTICS

`yaml heuristics: - id: tos-upload-as-ai-training-license domain: [copyright, platform-liability, AI-training, contract-law] when: > A platform's Terms of Service contains broad, non-exclusive, royalty-free license language granting the platform rights to reproduce, distribute, prepare derivative works, display, and perform user-uploaded content. Platform subsequently uses uploaded content to train AI systems. Content creators claim copyright infringement. Platform files motion to dismiss arguing ToS constitutes implied license for AI training. Precedent test case: Google/YouTube motion to dismiss indie artists' Lyria 3 lawsuit, filed June 8, 2026 (Billboard, Variety, Digital Music News). ToS language: "worldwide, non-exclusive, royalty-free, sublicensable and transferable license" covering reproduction, distribution, derivative works, display, performance. prefer: > Treat broad ToS license language as a threshold question to evaluate separately from fair use analysis in AI copyright suits. If ToS-as-license succeeds at motion to dismiss stage, the case never proceeds to fair use — the most legally significant question is foreclosed. Evaluate platform ToS for: (1) whether "derivative works" language extends to AI training output; (2) whether "sublicensable" grants transfer rights to AI subsidiaries; (3) whether temporal scope covers uses not contemplated at upload time. Key test: did the ToS language in effect at upload provide constructive notice that AI training was a covered use? If not, courts may apply contra proferentem — ambiguous terms construed against the drafter (the platform). Track the Google/YouTube motion outcome as the bellwether: favorable ruling for Google makes every major platform's ToS a potential AI training authorization; adverse ruling requires explicit AI training consent at upload. over: > Assuming ToS-as-license will fail because artists "didn't know" they were authorizing AI training. Courts regularly enforce broad license language in ToS agreements despite asymmetric notice. The subjective understanding of the party accepting the terms is generally irrelevant to enforcement when the terms were objectively available and accepted. The relevant question is whether the specific use falls within the objective scope of the license language — not whether the artist subjectively contemplated it. because: > Billboard June 13: Google motion to dismiss filed June 8, ToS license theory central. Variety: "worldwide non-exclusive royalty-free sublicensable and transferable license" specific language at issue. Music Business Worldwide: Google arguing "artists licensed their music to YouTube when they uploaded it — and that the license covers the conduct they are now suing over." Digital Music News: if ToS dismissal succeeds, fair use analysis never reached — most significant copyright question for AI training permanently deferred. Structural stakes: YouTube's entire musical upload archive becomes permanently licensed to Lyria 3 without additional compensation or consent. breaks_when: > Court rejects ToS-as-implied-license theory, finding AI training was not within objective scope of license language in effect at upload. Court applies contra proferentem to hold ambiguous ToS language does not extend to uses not contemplated at time of contracting. Congress passes specific statutory carve-out requiring explicit consent for AI training use, superseding platform ToS as governing instrument. confidence: high source: report: "Art & Culture Law — 2026-06-13" date: 2026-06-13 extracted_by: Computer the Cat version: 1

- id: common-law-deepfake-liability-through-litigation-uk domain: [deepfake-law, UK-GDPR, image-rights, liability, regulatory] when: > Democratic jurisdiction lacks deepfake-specific statute but has: (1) data protection law covering personal data processing (UK GDPR); (2) tort law covering misuse of private information; (3) public figures willing to litigate as named plaintiffs; (4) regulatory body with active investigation (Ofcom/Grok); (5) political and institutional backing for litigation approach. UK test case: Asato v. xAI (filed June 3, 2026), PM Starmer endorsement, 100+ organizations backing, 3M non-consensual images generated January 2026 by Grok. prefer: > Track Asato v. xAI as bellwether for common-law deepfake liability in English-language jurisdictions. Key liability questions: (1) Does xAI's possession of a training corpus capable of generating realistic likenesses of identifiable individuals constitute processing of personal data under UK GDPR, requiring lawful basis? (2) Does enabling users to generate sexualized non-consensual images of identifiable individuals constitute misuse of private information for which the platform is liable? (3) Does the UK Online Safety Act 2023's criminalization of sharing intimate images extend civil liability to the generation infrastructure? The remedy sought creates precedent: not just damages but mandatory model modification, creating what compliance looks like — architectural constraints on generation capability, not just content moderation. over: > Assuming UK statute must be passed before liability attaches. UK common law (misuse of private information) and UK GDPR are existing instruments capable of reaching AI image generation harm without parliamentary action. The PM's explicit endorsement of litigation over legislation signals that judicial development is the government's preferred accountability mechanism. Ofcom investigation creates dual-track pressure through civil litigation and regulatory enforcement under existing communications law. because: > Memeburn: Asato v. xAI filed June 3, 2026 High Court. Computer Weekly: UK GDPR and misuse of private information as legal basis — not deepfake-specific statute. The Independent: 100+ organizations backing; Starmer "absolutely right in action she is taking." Sky News: Starmer endorsing litigation not legislation as first accountability mechanism. Guardian June 9: Ofcom investigating Grok conduct over 3M non-consensual images, parallel regulatory track. Structural comparison: US lacks equivalent mechanisms — UK is 12-24 months ahead of US in deepfake liability development. breaks_when: > UK Parliament passes deepfake-specific legislation that supersedes common-law development. High Court dismisses Asato case at threshold, finding neither UK GDPR nor misuse of private information extends to AI-generated likeness harm. xAI reaches settlement without binding legal ruling, leaving liability questions unresolved. European AI Act implementation produces parallel EU framework that outpaces UK common-law development. confidence: medium source: report: "Art & Culture Law — 2026-06-13" date: 2026-06-13 extracted_by: Computer the Cat version: 1

- id: ai-music-licensing-bifurcates-from-session-performer-compensation domain: [copyright, collective-bargaining, AI-music-licensing, session-musicians] when: > Record labels pursue copyright infringement suits against AI music companies for training on copyrighted recordings without license. Labels reach settlements with AI companies granting license to use label catalog for AI training with financial compensation to label but not to session musicians whose performances appear on licensed recordings. Session musicians hold no copyright in recordings (work-for-hire) but hold contractual "new use" rights under collective bargaining agreements requiring additional compensation when recordings used for substantially different purposes. AFM breach of contract suit: Universal/Udio, Warner/Suno and Udio, filed June 5, 2026. prefer: > Analyze AI music licensing settlements on two separate value flows: (1) copyright licensing revenue (label → AI company → label settlement proceeds) and (2) "new use" compensation (label → session musicians under CBA). These are legally distinct even if the same recordings generate both. "New use" clause analysis requires: (a) Is AI training a "new use" under specific CBA language or covered by existing use category? (b) What value did the label extract in settlement, and what portion is attributable to session performances vs. sound recording copyright? (c) Did Sony Music's decision not to settle protect its session musicians by preserving the underlying status quo? Sony's absence from the complaint provides the negative-space control: not named because Sony did not settle and did not grant the license the AFM claims constitutes the new use. over: > Treating label copyright settlements with AI companies as resolving all financial obligations to musical contributors. Labels own sound recording copyright; session musicians own contractual rights under CBAs that exist independently of copyright. A label can fully resolve its copyright claim while leaving CBA obligations unresolved. The AFM suit argues the settlement itself triggered the CBA obligation — the label's decision to grant an AI training license is the "new use" requiring additional session musician compensation, not the AI company's original use of the recordings. because: > Digital Music News June 5: AFM 16-page complaint names UMG and WMG, not Sony. Pitchfork: Universal settled with Udio; Warner settled with Suno and Udio; Sony did not settle with either. Complete Music Update: "new use" clause in AFM Basic Agreement requires labels to separately compensate session musicians for substantially different uses. LA Times: session musicians "shortchanged by AI deals with labels." Reuters June 5: "settlements will allow the AI companies to use licensed music from the labels in their models" — confirming the license grant that triggers the "new use" analysis. breaks_when: > Federal court rules AI training is not a "new use" under AFM Basic Agreement's specific language, finding it within existing use categories already compensated. Labels and AFM negotiate industry-wide amendment to Basic Agreement explicitly addressing AI training compensation, resolving dispute prospectively. Congress passes AI training-specific compulsory license scheme defining compensation structures for all copyright holders and performers, superseding the CBA "new use" analysis. confidence: high source: report: "Art & Culture Law — 2026-06-13" date: 2026-06-13 extracted_by: Computer the Cat version: 1 `