Blog

The AI Creator Compensation and Licensing Act (AICCLA)

EXECUTIVE SUMMARY

Independent creators are shut out of the AI training data economy. Foundation models train on billions of copyrighted works, yet the emerging licensing market serves only institutional licensors: News Corp secured $250 million from OpenAI [1]; Shutterstock generated over $100 million in AI licensing revenue [2]. Independent creators receive nothing. Over 70 copyright lawsuits have produced conflicting rulings [3], and the $1.5 billion Bartz v. Anthropic settlement demonstrates the cost of case-by-case adjudication [4]. This brief proposes the AI Creator Compensation and Licensing Act (AICCLA), establishing The AI Licensing Collective (ALC): a single nonprofit that manages blanket licenses for AI training data, a statutory royalty floor set by the Copyright Royalty Board, a federal registry linked to the CLEAR Act database, and a safe harbor that replaces litigation uncertainty with predictable costs.

PROBLEM STATEMENT

A Two-Tier Market That Excludes Independent Creators

For any individual work, the cost of identifying, contacting, and negotiating with an AI developer exceeds the expected royalty, making bilateral licensing economically irrational for all but the largest catalogs. OpenAI has signed 18 publisher deals. None involve independent creators. CISAC and PMP Strategy project cumulative creator revenue losses of $24 billion through 2028 across music and audiovisual sectors [5]. The Copyright Office’s May 2025 report acknowledged this structural gap but deferred to Congress [6].

Judicial Incoherence Demands a Legislative Framework

Federal courts have produced conflicting fair use rulings. In Bartz v. Anthropic (N.D. Cal. 2025), training on legally acquired works was held “quintessentially transformative,” but training on pirated sources constituted infringement, producing a $1.5 billion settlement [4]. In Kadrey v. Meta (N.D. Cal. 2025), the court found LLM training on copyrighted works to be fair use, including material from shadow libraries [7]. In Thomson Reuters v. Ross (D. Del. 2025), fair use was rejected for a non-generative AI system [8]. Over 70 lawsuits are now pending [3]. Under 17 U.S.C. §504, willful infringement carries statutory damages of up to $150,000 per work, creating theoretical exposure of $150 billion for a model trained on one million registered works.

PROPOSED POLICY: THE AI CREATOR COMPENSATION AND LICENSING ACT (AICCLA)

Pillar 1: The AI Licensing Collective with Extended Collective Licensing

Congress directs the Copyright Office to designate a single nonprofit collective, The AI Licensing Collective (ALC), to manage blanket licenses for AI training data. The ALC operates sector divisions by content modality (initially text, image, and audio/music, expandable), each with independent royalty pools. Works under OSI-approved open-source licenses are excluded.

Under extended collective licensing (ECL), the ALC negotiates blanket license terms with AI developers. ECL coverage extends to all copyrighted works by operation of law, securing opt-out rights regardless of registration status. Any creator may withdraw works from future training at no cost. Registration under 17 U.S.C. §408 is required to receive royalty distributions.

The CRB sets a statutory royalty floor as a minimum guarantee. The ALC negotiates above this baseline through market bargaining. The MMA’s blanket mechanical license provides domestic precedent; the Supreme Court confirmed blanket licensing’s antitrust compatibility in BMI v. CBS [9]. The guaranteed opt-out, combined with escrow compensation, satisfies due process requirements. Because no existing investment-backed expectation in AI training licensing exists for most creators, ECL does not affect a regulatory taking under Penn Central standards [10].

Pillar 2: Tiered Revenue Structure

The CRB determines “AI service revenue” through notice-and-comment rulemaking, applying Section 115’s “service revenue” methodology including proportional allocation for bundled offerings [11]. Revenue thresholds defining each tier are set by the CRB; as an illustrative framework, Tier 1 applies below $10 million in annual AI service revenue, Tier 2 between $10 million and $1 billion, and Tier 3 above $1 billion. Rates follow four tiers:

Royalty obligations attach at commercial deployment in the United States regardless of training location, including developers who commercially deploy open-weight models. Models trained on outputs generated by covered AI systems inherit disclosure obligations; the CRB may extend royalty treatment to derivative training through rulemaking.

Existing bilateral agreements between AI developers and institutional licensors remain valid. Bilateral payments satisfy ALC obligations for those works, incentivizing participation: institutional licensors retain deal premiums while gaining ALC coverage for unlicensed works.

Pillar 3: Federal Registry and Distribution

The AICCLA directs the Copyright Office to establish a Federal AI Training Data Registry, a free platform enabling any registered copyright owner to claim works and set licensing preferences. The CLEAR Act (S.3813), with bipartisan support in the 119th Congress [12], requires AI developers to disclose training data. The ALC cross-references these disclosures to

notify affected owners and requires machine-readable submissions aligned with NIST provenance standards [13].

Distribution follows a two-factor model within each sector: a base allocation proportional to training data volume, adjusted by a market-value multiplier that the CRB reassesses periodically. Individual contribution tracking remains infeasible at scale; this proxy ensures both the scale and demonstrated value of each work inform compensation. Under illustrative rates, the annual royalty pool would exceed $2.5 billion. Even creators in the lowest distribution tier would receive over $1,000 annually, compared to $0 today, alongside legal rights to opt out of future training. Bloomberg Intelligence projects the generative AI market to reach $1.3 trillion by 2032 [14]; as revenue grows, the pool scales proportionally.

Royalties for non-participating owners are held in escrow for five years, consistent with the Copyright Office’s MLC recommendation [15]. After five years, unclaimed funds are allocated to IP registration subsidies for individual copyright owners. The ALC subsidizes registration at no upfront cost to creators.

Pillar 4: Safe Harbor for Participating AI Developers

AI developers that execute a blanket license with the ALC receive a statutory safe harbor:

• Immunity from statutory damages for covered works, following the MMA §115(d)(1) model [11]
• Injunctive relief limited to monetary remedies
• Presumption of good faith, precluding willfulness findings (17 U.S.C. §504)

For non-participating developers, the Act raises the statutory damage floor from $750 to $2,500 per work (§504(c)(1)) for developers who decline available blanket licenses. With over 70 pending lawsuits and $3 billion in additional claims [16], cumulative litigation exposure far exceeds predictable licensing costs.

IMPLEMENTATION CONSIDERATIONS

Governance. A board ensuring balanced representation: illustratively, 5 individual copyright owners, 3 institutional licensors, 3 AI developers, and 2 public interest experts. No single stakeholder group holds a majority. Rate-setting decisions require a two-thirds supermajority, preventing any two groups from overriding creator interests.

Funding. AI developers fund operations through an administrative assessment set by the CRB, separate from royalty payments. The MLC operates on a $39 million annual assessment [15]. Scaling to three sectors, the ALC’s estimated annual operating cost is $100–150 million, funded through CRB-set assessments. Initial startup costs are funded through a congressional appropriation during a two-year establishment period.

Political feasibility. The AICCLA builds on bipartisan momentum: the CLEAR Act has cross party sponsors, the MMA passed unanimously, and creators’ guilds and AI developers alike prefer predictable rules to litigation uncertainty.

Phased rollout modeled on MLC precedent. The MLC achieved full operation within 27 months for a single sector [15]. Extrapolating to three sectors with sequential launches, full ALC operation is projected within 4–5 years: disclosure requirements first, then registry launch, then blanket licensing in each sector.

ANTICIPATED OBJECTIONS AND RESPONSES

“Fair use already permits AI training.” Courts disagree. The AICCLA does not modify Section 107. Fair use remains available as a defense. The Act creates a parallel pathway: developers who prefer certainty over litigation obtain it through blanket licensing, following Congress’s practice under Sections 111, 114, and 115 [11].

“Open-source models and offshore training evade this framework.” Royalty obligations attach at commercial deployment in the United States, not at training. The deploying entity bears obligations on its own service revenue, consistent with Section 115’s distribution-based trigger [11].

“Individual contribution tracking is technically infeasible.” Correct. The ALC uses training data volume and market value as distribution proxies, paralleling how music CMOs use play counts rather than measuring each song’s contribution to listener satisfaction. As attribution technology matures, the CRB adjusts the formula.

“This will slow AI innovation.” Over 70 lawsuits and $1.5 billion in settlements [4] create far greater uncertainty than predictable licensing. The Act attaches compensation to commercial use without restricting what AI systems learn or generate, following Sections 111 and 114. The tiered structure ensures startups and academic institutions face no burden.

“The Copyright Office did not recommend collective licensing.” Correct. The Office preferred market solutions and set a threshold: collective alternatives when markets prove inadequate [6]. Eighteen months later, eighteen publisher deals compensate institutional licensors while independent creators receive nothing. That threshold is met.

REFERENCES

1. News Corp-OpenAI Partnership, $250M multi-year agreement (2024).
2. Shutterstock AI licensing revenue, $104M (2023). PetaPixel, June 2024.
3. Copyright Alliance, “AI Copyright Lawsuit Developments in 2025: A Year in Review” (Jan. 2026).
4. Bartz v. Anthropic, No. 3:24-cv-05417 (N.D. Cal. 2025); Settlement approved Aug. 2025 ($1.5B).
5. CISAC & PMP Strategy, Global Economic Study: The Impact of Generative AI on Creators (2025).
6. U.S. Copyright Office, Copyright and Artificial Intelligence, Part 3: Generative AI Training (May 2025).
7. Kadrey v. Meta Platforms, No. 3:23-cv-03417 (N.D. Cal. 2025).
8. Thomson Reuters v. Ross Intelligence, No. 1:20-cv-613 (D. Del. 2025).
9. BMI v. CBS, 441 U.S. 1 (1979).
10. Penn Central Transportation Co. v. New York City, 438 U.S. 104 (1978).
11. Copyright Act §§ 111, 114, 115, 504, 17 U.S.C.; Music Modernization Act, Pub. L. No. 115-264 (2018).
12. CLEAR Act, S.3813, 119th Cong. (2026).
13. NIST, AI Risk Management Framework 1.0 (AI 100-1, 2023); NIST, Generative AI Profile (AI 600-1, 2024).
14. Bloomberg Intelligence, “Generative AI to Become a $1.3 Trillion Market by 2032” (June 2023).
15. Mechanical Licensing Collective, Annual Report and Milestones (2025); Uniform Unclaimed Property Act (2016).
16. Concord Music Group v. Anthropic, No. 5:26-cv-00880 (N.D. Cal. 2026) ($3B+ claimed).
17. ASCAP, Annual Financial Results (2024): $1.7B distributed to over 1M members.
18. EU Directive 2019/790, Art. 12 (Extended Collective Licensing with Extended Effect).
19. C2PA, Content Credentials Technical Specification v2.2 (2025).
20. Copyright Royalty Board, 37 C.F.R. Part 385.

AI TOOL DISCLOSURE

This policy brief was developed with the assistance of Claude (Anthropic) for background research, and data compilation. The AI tool was used to gather and organize publicly available legal, economic, and legislative information.

Published May 2, 2026

← Back to blog home