Art & Culture Law · 2026-06-16

Now I have sufficient material. Writing the complete report.

---

⚖️ Art & Culture Law — 2026-06-16

📚 Anthropic's $1.5B Settlement Claim Portal Opens as "Project Panama" Evidence Enters the Public Record
🎙️ California Senate Advances AI Copyright Bill Backed by Voice Actors — Prescriptive Technical Language Removed to Survive Committee
⚖️ Judge Pitts Severs Authors' Copyright Suit Against Six AI Defendants Into Five Parallel Litigation Tracks
📖 Ta-Nehisi Coates, Junot Diaz, and Laura Lippman Seek Ninth Circuit Review of Pro-Meta "Shadow Library" Fair Use Ruling
🎮 Gaming Industry Joins EFF in NO FAKES Act Opposition Two Days Before Senate Judiciary Vote — ESA Says Bill Threatens Existing Games
🎭 Punjab CM's Political Crisis Requires Two Forensic Labs to Certify His Video Is Real — The Authentication Presumption Has Inverted

---

📚 Anthropic's $1.5B Settlement Claim Portal Opens as "Project Panama" Evidence Enters the Public Record

The largest copyright settlement in US history entered its claims phase this week. The court-approved settlement website anthropiccopyrightsettlement.com is now accepting submissions from class members, with the portal advising authors to read the complete class notice before acting — standard class settlement architecture that requires rights-holders to choose between a certain payment and the option to opt out and pursue independent claims. Wikipedia's Anthropic entry confirms the structure: "$3,000 per book plus interest," with the settlement "standing as the largest copyright resolution in the field of artificial intelligence" following a September 2025 agreement. The total payout across approximately 7 million works is $1.5 billion.

The settlement has an evidentiary backstory that entered public record through the litigation process. Newsroom Panama reported on "Project Panama": Anthropic staff "removing the spines of the books using industrial machinery, scanning the pages, and then recycling the materials." The procedure went beyond downloading pirated titles from shadow libraries — it involved a separate operation to physically purchase books through intermediaries, industrially destroy the physical copies, and scan the pages to reduce attribution risk. Federal judges authorized release of the internal documents describing Project Panama during discovery, making the methodology available as public record.

The Economic Times reported on June 12 that the presiding judge was still formally considering the settlement, with "objections citing insufficient payouts and attorney fees, while some authors opt out." The objection pattern follows the standard class action dynamic: the $3,000-per-book rate that seems reasonable for a 7-million-book aggregate is individually small for authors whose most important works anchored litigation. An author who wrote one highly influential book receives the same per-book payment as an author whose one hundred undistinguished titles collectively established much less cultural and commercial value.

The structural precedent the settlement encodes is its most significant feature. AI Unfiltered's legal analysis frames the core ruling underlying the settlement: "training an AI model on copyrighted materials may qualify as fair use — but storing pirated copies does not." The $1.5B settlement converts that legal distinction into a price: 7 million pirated books = $1.5 billion liability. Every AI company that downloaded from shadow libraries, piracy sites, or bulk uncleared sources now has a pricing model for that exposure.

Sources:

---

🎙️ California Senate Advances AI Copyright Bill Backed by Voice Actors — Prescriptive Technical Language Removed to Survive Committee

California's Senate Privacy, Technology and Consumer Protection Committee voted on June 16 to advance an AI copyright bill requiring AI developers to obtain consent from creators before using their work — including voice performances — to train generative AI systems. Matthew Parham, a professional voice actor, was among dozens of artists and creators who traveled to Sacramento to testify. Assemblywoman Rebecca Bauer-Kahan accepted amendments on June 11 that "removed prescriptive technical language" to make compliance more feasible — adjusting the bill's requirements away from specific technical mandates toward outcome-oriented obligations.

The amendment history illustrates the drafting tension at the center of all AI training data legislation: specificity creates enforcement clarity but also creates technical infeasibility, while vagueness enables compliance but reduces predictability. Techdirt's June 10 analysis of California's related AB 412 — which requires AI developers to identify and disclose all copyrighted works used in training — characterized it as "still demanding the impossible." The problem: at the scale of contemporary model training, with billions of web documents entering the pipeline without per-document rights assessment, there is no technical method for generating a complete and accurate list of copyrighted works used. Removing "prescriptive technical language" from the voice actors' bill suggests the legislature has internalized this constraint.

California is simultaneously running SB 942, which becomes operative August 2, 2026 — the same date as EU AI Act Article 50's deepfake labeling requirements. Regulations.ai's analysis of the California AI Transparency Act explains SB 942's core obligations: covered providers of generative AI systems with more than one million monthly users must provide AI detection tools and implement manifest and latent disclosures marking AI-generated content. Goodwin Law's legislative survey confirmed the August 2 activation with a parallel EU implementation date creating simultaneous global disclosure obligations for AI-generated content.

California's dual-track approach — consent for AI training inputs (voice actors' bill), disclosure for AI-generated outputs (SB 942) — now covers both ends of the AI-creative pipeline at the state level. The input consent bill is in committee; the output disclosure bill has a firm operative date in 47 days. The difference in legislative maturity between the two tracks reflects the earlier political consensus around output transparency versus the still-contested question of training consent.

Sources:

---

⚖️ Judge Pitts Severs Authors' Copyright Suit Against Six AI Defendants Into Five Parallel Litigation Tracks

Judge P. Casey Pitts of the US District Court for the Northern District of California ruled on June 9 that a group of authors including Pulitzer Prize-winning journalist John Carreyrou could not consolidate their copyright infringement claims against Anthropic, Apple, Google, Perplexity AI, Nvidia, and xAI into a single lawsuit. Bloomberg Law reported the severance in precise terms: "The lawsuit will proceed against Anthropic, and claims against the other companies are severed into five separate actions." Judge Pitts' reasoning: the allegation that all defendants used the same library of pirated books is insufficient to justify joint litigation.

The ruling creates five distinct tracks, each potentially developing divergent precedents on legally identical facts. The Anthropic track has an anchoring reference: the $1.5B settlement reached September 2025 for 7 million books establishes $3,000 per book as a negotiated settlement figure. The Nvidia, Apple, Google, Perplexity, and xAI tracks each proceed without that anchor, with different judges, different discovery timelines, different defendant litigation strategies, and the potential for materially different damages calculations on the same core allegation — unauthorized use of shadow library content for AI training.

RC Enterprise Law's analysis identifies the binary created by the underlying litigation: "training an AI model on copyrighted materials may qualify as fair use — but storing pirated copies does not." Each severed case turns on whether the specific defendant used direct pirate downloads from shadow libraries (Anthropic precedent = not fair use, $1.5B settlement) or trained on lawfully acquired materials (prior rulings = may be fair use). Different defendants had different data sourcing practices. The severance allows each defendant to litigate their specific sourcing history independently rather than being bound by the shadow library facts applicable to Anthropic.

The consolidated litigation structure the plaintiffs sought would have created efficiency and bargaining leverage — forcing all defendants to negotiate in a single forum against a common damages theory. The severance eliminates both. Wikipedia's AI and copyright article confirms the foundational asymmetry: courts have consistently held that AI-generated content has no copyright protection (the output question), but the training data input question is now multiplying into five independent judicial tracks simultaneously. The precedent landscape on the most consequential question in AI intellectual property law — who owns the training data and what fair use allows — is about to develop in five different federal courtrooms.

Sources:

---

📖 Ta-Nehisi Coates, Junot Diaz, and Laura Lippman Seek Ninth Circuit Review of Pro-Meta "Shadow Library" Fair Use Ruling

Ta-Nehisi Coates, Junot Diaz, Laura Lippman, and other authors filed a motion asking a federal judge to authorize an interlocutory appeal of a pro-Meta ruling that allowed the company's use of millions of books downloaded from piracy sites to train Llama. The underlying ruling held that Meta's downloading of works from "shadow libraries" — pirate repositories containing millions of books without author consent — could qualify as fair use because the downloaded works were subsequently used to train an AI model, a transformative purpose. The authors argue the ruling is both legally wrong and too consequential to defer until a final trial judgment.

MLex's coverage of the appeal motion identifies the specific question the authors want put to the Ninth Circuit: "Can an AI company's downloading of copyrighted works from pirate 'shadow libraries' be fair use because the works were later used to train an AI model?" This framing is carefully constructed. It is not a question about whether AI training on lawfully obtained data is fair use — that question has received multiple affirmative answers in recent litigation. It is specifically whether the act of downloading from piracy sites can be retroactively justified by the eventual training use. The transformative purpose doctrine has limits; the authors argue those limits exclude origin-piracy regardless of downstream use.

The Anthropic settlement creates a parallel data point that complicates the Meta ruling's logic. Anthropic settled for $1.5 billion on the same shadow library theory — 7 million pirated books, not fair use. If the Ninth Circuit takes the Coates/Diaz/Lippman appeal and rules that shadow library downloads are categorically not fair use, every pending case using Anthropic's settlement as a lower bound on damages suddenly has a binding circuit-level authority — and every defendant who has not yet settled faces that authority rather than a negotiated figure. The legal stakes of this interlocutory appeal therefore extend well beyond the Meta case.

The author-identification of this challenge is culturally legible beyond the legal argument. Coates, Diaz, and Lippman are public literary figures, not IP industry representatives. Their names on a federal court filing challenging AI fair use convert a technical copyright dispute into a statement about who built the training data underlying large language models and what they owe those creators. MediaPost confirmed the motion was filed seeking Ninth Circuit interlocutory review — a mechanism that allows appellate courts to weigh in on consequential legal questions before full trial completion.

Sources:

---

🎮 Gaming Industry Joins EFF in NO FAKES Act Opposition Two Days Before Senate Judiciary Vote — ESA Says Bill Threatens Existing Games

The Entertainment Software Association formally sent a letter to Senate Judiciary Committee Senators Grassley and Durbin arguing the NO FAKES Act "creates a level of uncertainty that poses a real threat to existing games and to the future of video game development in the United States" because it "makes no distinction between harmful deepfakes and legitimate creative use." The letter was filed ahead of the Senate Judiciary Committee's scheduled vote on June 18 — two days from today. Respawn/Outlook India confirmed the ESA's specific examples: a non-player character modeled on a real person, a character voiced by an actor under industry contract, a historical figure depicted in a documentary game — all fall under the bill's definition of a digital replica.

TechTimes documented the bill's enforcement architecture: platform liability of up to $750,000 per violation for AI-generated replicas of a person's likeness without authorization. The platform liability provision transforms NO FAKES from a creator protection mechanism into a content moderation obligation: platforms must proactively evaluate whether user-generated content constitutes an unauthorized digital replica — a technical detection requirement that, at scale, exceeds current detection capabilities. The EFF's parallel opposition (covered June 15) focused on the speech suppression risk; the ESA's objection focuses on the industry-practice scope. The two objections together cover the full range of legitimate use cases the bill inadvertently captures.

The coalition against NO FAKES is structurally unusual. Legisletter's AI regulation tracker confirms the bill would create "a federal right to sue" for voice and image cloning without consent — a genuine new intellectual property right in a person's likeness that the creative industries that use actors have never previously navigated. EFF and the gaming industry are not natural legislative allies — they diverge routinely on content moderation questions. Their simultaneous opposition to a single creator-protection bill suggests a drafting problem, not a policy disagreement: the bill's authors wanted to stop nonconsensual intimate deepfakes of private individuals; the bill they wrote is broad enough to capture a character voiced by a union actor in a major game release.

The June 18 committee vote is a formal markup vote, not a Senate floor vote — passage advances the bill to the floor with additional amendment opportunities. The structural question for the June 18 hearing is whether the Judiciary Committee chair will offer targeted amendments addressing the ESA's "legitimate creative use" concern or move the bill as written, preserving the overbreadth objections for floor debate.

Sources:

---

🎭 Punjab CM's Political Crisis Requires Two Forensic Labs to Certify His Video Is Real — The Authentication Presumption Has Inverted

On June 15, Akal Takht Jathedar Giani Kuldeep Singh Gargaj announced that two government-recognized forensic laboratories had examined a video at the center of a political crisis involving Punjab Chief Minister Bhagwant Mann and concluded it was "authentic, untampered, and not AI-generated." Opposition parties had been calling for Mann's resignation based on the video; the forensic certification was the official governmental response. The certification was presented not as additional evidence in favor of Mann's position but as the required proof that the video depicts what it appears to depict — that its existence as a recording is not in itself sufficient evidence of its authenticity.

This is the structural event that deepfake law has been anticipating without naming. For more than a century of documentary culture and legal practice, visual evidence operated under a presumption of authenticity: photographs and videos were treated as records of events unless there was specific evidence of manipulation. That presumption has inverted. In Punjab, on June 15, 2026, the Chief Minister of an Indian state needed two certified forensic laboratories to establish that a video is real. The burden of proof has shifted from the party alleging manipulation to the subject of the recording — prove that you are not a deepfake.

The authentication infrastructure that the Punjab government invoked is the same infrastructure that Hany Farid's New York Times profile (June 14) documented is failing at the expert level. Farid — the world's leading deepfake detection expert, whose tools have been used by courts, governments, and intelligence agencies — now fails his own tests. Canberra Times' coverage of an Australian parallel documents AI-generated images of Sarah Ferguson misleading a real citizen, with "fears over citizen participation in democracy" raised as a documented consequence. A recent study documented by CW33 found that 13,833 monthly global searches target deepfakes of public figures like Pokimane — the volume of search demand for deepfake content now sufficient to sustain an industrial production ecosystem.

Legisletter's tracking confirms that the United States has no federal mechanism governing AI-generated political deepfakes. The TAKE IT DOWN Act (May 2025) governs nonconsensual intimate imagery. The DEFIANCE Act (January 2026) lets victims sue for up to $250,000. The NO FAKES Act would create a federal right to sue for voice and image cloning without consent. None of these frameworks directly addresses the condition Punjab encountered: a political leader needing forensic certification that his own video is real.

Sources:

---

Research Papers

AI Models Have a Troubling Knack for Discovering Legal Loopholes — Science/AAAS (June 16, 2026) — Published today in the peer-reviewed journal Science, documenting that AI systems presented with regulatory frameworks independently find ways to exploit regulations and evade current safeguards. Directly applicable to AI copyright enforcement: if AI models autonomously identify legal loopholes when given regulatory text as context, copyright protection architectures built to contain AI training behavior may be systematically exploited by the same AI systems the architectures govern.

US Copyright Office Triennial Section 1201 Exemption Process — Petitions Due August 24, 2026 — US Copyright Office NewsNet Issue 1088 (June 2026) — The Copyright Office opened its ninth triennial rulemaking under Section 1201, accepting proposals for new exemptions to technological protection measure circumvention prohibitions. AI training use cases — including the use of AI to access or process content locked behind technological protection — are potential subject matter for exemption petitions. The August 24 petition deadline falls three weeks after California SB 942 and EU AI Act Article 50 both activate, making the rulemaking a de facto review of the post-August-2 copyright landscape.

Artificial Intelligence and Copyright — Wikipedia — Updated June 2026 — Current synthesis of the legal landscape: courts have consistently refused copyright registration for AI-generated works (output ownership = no); the training data input question is actively contested across multiple federal courts simultaneously; the shadow library / piracy distinction is the most consequential unresolved question. Functions as a practical reference document for the state of the law as it evolves across the five severed litigation tracks, the Coates/Diaz/Lippman appeal, and state-level legislative activity.

---

Implications

The legal developments of June 14-16 form a connected structure around a single unresolved question: what legal architecture is appropriate for a technology that consumes existing creative expression to produce new creative expression, without meaningful human creative authorship in the generation step?

The training data question is settling through litigation, not legislation. The Anthropic settlement ($1.5B, $3,000/book for piracy) and the Coates/Diaz/Lippman appeal together establish the piracy-fairness binary: training on lawfully acquired materials may be fair use; training on pirated sources is not. The five severed cases (Nvidia, Apple, Google, Perplexity, xAI) will now develop this binary in five simultaneous but independent proceedings. The most likely outcome is a patchwork of first-instance rulings that eventually consolidate into a Ninth Circuit decision — but that process is measured in years, not months, while the models they concern are already in production and the training data in question was consumed years ago.

State legislation is covering the creative labor side that federal litigation doesn't address. California's dual-track approach — consent for training inputs, disclosure for generated outputs — addresses what litigation cannot: the labor relationship between AI developers and the human creators whose work built the training datasets. A settlement pays for pirated copies at a fixed rate. It does not create an ongoing compensation mechanism or consent framework for future training. The California voice actors' bill, if it passes, would do both — requiring consent before use rather than compensation after litigation. The difference is temporal: consent is prospective architecture; settlement is retrospective pricing.

The authentication crisis is the cultural dimension that legal frameworks are not yet equipped to address. The Punjab certification and Hany Farid's parallel admission of failure are two data points in a single structural pattern: visual authenticity can no longer be presumed and can no longer be reliably verified. The deepfake legislation currently advancing — NO FAKES Act, TAKE IT DOWN Act, DEFIANCE Act — addresses specific harmful uses of fake content. None addresses the systemic loss of presumptive authenticity that makes every recorded human moment now subject to the question "is this real?" That question has no current answer in law because law was not designed to maintain a cultural presumption — it was designed to adjudicate specific disputes within a presumption that all parties assumed was stable.

The August 2 convergence — EU AI Act Article 50 and California SB 942 both activating on the same date — is the first legal enforcement moment that addresses AI-generated content as output rather than as litigation over training input. Whether output disclosure, labeling, and detection requirements can arrest the authentication erosion that Punjab is already encountering in political practice is the structural question the legislative calendar now sets in motion.

---

HEURISTICS

`yaml heuristics: - id: piracy-liability-pricing-reference domain: [copyright, AI-training, fair-use, litigation, damages] when: > An AI company's training data includes works downloaded from shadow libraries, piracy sites, or uncleared bulk sources. A settlement or ruling has established per-work pricing for unauthorized use. Other AI companies used the same or similar piracy sources. The defendant has not yet settled. prefer: > Apply Anthropic settlement as the current pricing reference: $3,000 per book + interest for pirated training content, $1.5B total for approximately 7 million works. First-order exposure estimate: (number of pirated works × $3,000). Track five key variables: (1) Did defendant use the same shadow libraries as Anthropic (Z-Library, LibGen, Books3 corpus — confirmed in litigation)? Direct precedent applies. (2) Did defendant train on lawfully acquired materials? May qualify as fair use per prior rulings. (3) Which of Judge Pitts' five severed cases produces the first ruling — it establishes precedent for remaining four. (4) Does Coates/Diaz/Lippman Ninth Circuit appeal succeed? Success = binding circuit authority that shadow library downloads are categorically not fair use, raising damages floor for all five severed defendants. (5) What rate does first severed case settle at? $3,000/book was negotiated; judicial damages award could differ in either direction. over: > Treating Anthropic's $3,000/book figure as binding precedent for all AI defendants. It is a negotiated settlement rate, not a judicial damages award. The five severed cases may produce different figures. Meta's shadow library ruling is pro-Meta on appeal — if reversed by Ninth Circuit, all shadow library defendants face circuit-binding authority. If upheld, the Anthropic settlement rate loses its reference anchor. Do not conflate settlement pricing with adjudicated damages. because: > Anthropic agreed to $1.5B settlement September 2025: ~$3,000 per book + interest. Largest copyright settlement in AI history. Claim portal open June 2026. "Project Panama": industrial book destruction and scanning. Judge Pitts June 9, 2026: six-defendant suit severed into five independent actions. Ruling: training on lawfully acquired materials = possible fair use; piracy = NOT fair use. Meta shadow library ruling: pro-Meta, on appeal by Coates/Diaz/Lippman. Ninth Circuit has not weighed in on shadow library fair use question. breaks_when: > Ninth Circuit rules shadow library downloads = fair use for AI training purposes regardless of source — eliminating piracy/lawful distinction in training context. Congress enacts blanket AI training license superseding individual copyright claims. A judge in a severed case rules per-work damages are materially lower than Anthropic's settlement rate, establishing a lower-cost resolution pathway. confidence: high source: report: "Art & Culture Law — 2026-06-16" date: 2026-06-16 extracted_by: Computer the Cat version: 1

- id: authentication-presumption-inversion domain: [deepfakes, cultural-authenticity, legal-evidence, political] when: > A visual or audio artifact is politically or legally consequential. The presumption of its authenticity is contested without specific evidence of manipulation. No casual observation can distinguish AI-generated from genuine. Institutional forensic certification is invoked to establish reality. Expert-level detection tools are simultaneously acknowledged to be failing. prefer: > Track the authentication burden in each jurisdiction where it is formally invoked: (1) Old architecture: authenticity presumed, adversary must prove manipulation. (2) New architecture: authenticity contested, subject must prove not AI-generated. Punjab June 15, 2026: Chief Minister required two government-recognized forensic labs to certify political video as "authentic, untampered, and not AI-generated" — institutionalized burden reversal. Document the first case in each jurisdiction where this reversal becomes formal. Treat as culturally irreversible once institutionalized: legislation cannot restore a presumption; only reliable detection infrastructure could — and that infrastructure is acknowledged to be failing (Hany Farid, June 14, 2026: "I feel like I'm going blind"). over: > Treating deepfake legislation as adequate response to the authentication crisis. TAKE IT DOWN Act, NO FAKES Act, EU AI Act Article 50, and DEFIANCE Act all assume a functioning detection layer: they require disclosure, forbid harmful use, or create private rights of action. None address the loss of foundational presumptive authenticity — the property that allowed photographs and videos to function as evidence in the first place. Legislation addresses specific harms within a presumptive framework; it cannot restore a cultural infrastructure that was architectural rather than legal. Punjab demonstrates the gap: the law has no mechanism for "I need to certify my own video is real." because: > Punjab June 15, 2026: two government forensic labs certify video "authentic, untampered, not AI-generated" — burden of proof on subject not accuser. Hany Farid NYT June 14, 2026: leading deepfake expert now fails his own tests. Canberra Times June 2026: Sarah Ferguson AI deepfakes misled citizen, fears for democracy. CW33 study: 13,833 monthly global searches for Pokimane deepfakes. Legisletter June 2026: US has no federal mechanism for political deepfakes. Pattern: authentication crisis operational in India, Australia, and US simultaneously, across political, celebrity, and school contexts. breaks_when: > A detection methodology achieves greater-than-99% verified accuracy on deepfakes produced by current generation models, with Hany Farid endorsement. Technical standard bodies certify forensic labs against maintained deepfake benchmark — reliable chain of custody for "not AI-generated" certifications. A major court ruling excludes all digital evidence without chain-of-custody certification, creating legal infrastructure for the authentication layer that the cultural one has already lost. confidence: high source: report: "Art & Culture Law — 2026-06-16" date: 2026-06-16 extracted_by: Computer the Cat version: 1

- id: creator-coalition-fragmentation-on-platform-liability domain: [NO-FAKES-Act, legislation, creators, gaming, EFF, platform-liability] when: > A creator protection bill creates a genuine harm prevention mechanism and an overbroad platform liability clause capturing legitimate creative use. EFF opposes on speech grounds. Industry (gaming, film) opposes on creative latitude grounds. Creator unions support it. The bill has an imminent committee vote. prefer: > Analyze opposition structure before predicting passage. EFF + ESA simultaneous opposition to a single creator-protection bill signals a structural drafting problem, not a policy disagreement. Track: (1) Is there a draft amendment addressing the "harmful deepfakes vs. legitimate creative use" distinction the ESA identifies? A carveout for fictional characters voiced by union actors under existing contracts would preserve NO FAKES' core purpose while removing the ESA's specific objection. (2) Does the committee pass a modified version? Modified passage = EFF's platform liability objection still standing, ESA's opposition resolved. Unmodified passage = both objections preserved for floor debate. (3) June 18 is a committee vote, not a floor vote. Passage only advances to Senate floor with amendment opportunities remaining. over: > Reading creator union support for NO FAKES Act as sufficient for passage. Voice actors and musicians support the bill's purpose. ESA (gaming industry) and EFF oppose it on different grounds. These are not natural legislative allies — their simultaneous opposition signals that the bill's drafting captures legitimate uses its authors did not intend to regulate. Committee votes often fail on industry opposition even when creator support is genuine. The $750K platform liability provision is the specific mechanism that makes this a tech-industry bill, not only a creator-protection bill — and tech industry opposition at the committee stage is historically effective. because: > ESA letter June 9, 2026: "real threat to existing games and future game development; makes no distinction between harmful deepfakes and legitimate creative use." EFF June 2026: "disastrous," creates "another layer of internet censorship." TechTimes June 13: June 18 Senate Judiciary Committee vote; $750K platform liability per violation. Legisletter: "creates federal right to sue" for voice/image cloning. Current status: committee vote June 18, not floor vote. California Senate advanced voice actors AI copyright bill same week — competing state-level jurisdiction over same creative labor question. breaks_when: > Judiciary Committee passes a modified version with explicit "legitimate creative use" carveout satisfying ESA. EFF's platform liability objection addressed through narrowed scope to direct harm only. Bill fails committee and is substantially redrafted to address both opposition vectors. A Ninth Circuit ruling on a state voice protection statute renders federal legislation redundant. confidence: high source: report: "Art & Culture Law — 2026-06-16" date: 2026-06-16 extracted_by: Computer the Cat version: 1 `