On October 16, 2026, Elena Izyumenko, Assistant Professor of Intellectual Property Law at the University of Amsterdam's Institute for Information Law (IViR), posted a 15-page paper on SSRN arguing that the European Union's current approach to paying for AI training cannot be assumed to protect the people who actually wrote, drew, composed or performed the works - because the money it produces flows to whoever owns the copyright, and that is frequently a publisher, label or studio rather than the creator.

In Short

A law professor in Amsterdam wrote a paper asking whether AI companies paying for the books, songs and pictures they learn from actually helps the people who made them. Her answer is that under EU rules the money usually goes to whoever owns the copyright, which is often a big company, so calling it a "human right" of creators can be misleading. She also checks the proposed fixes - a general AI tax for culture, or a legal right to a small payment for each creator - and finds that each one has its own weak spot.

A paper about who gets paid, not whether

The paper is titled "A Human Right to AI Remuneration? What Human Rights Law Tells Us Copyright Owes Creators." According to its SSRN abstract page, it was written and posted on October 16, 2026, carries SSRN id 7621139, and is licensed under Creative Commons Attribution-NonCommercial 4.0. When the page was captured, it showed 55 downloads and 90 abstract views. Izyumenko promoted it on LinkedIn the following day.

Its opening sentence sets the problem plainly. "AI remuneration may be defended in the name of creators' human rights while leaving creators themselves with little or nothing," according to the abstract.

The target is a specific line of argument that has gained ground in Brussels. The paper quotes the European Parliament Resolution of 10 March 2026 on copyright and generative AI, which states that respect for the "fundamental right" to intellectual property "must be guaranteed at all stages of the digital transformation and of the development of generative artificial intelligence," and that "the widespread violation of copyright rules by GenAI providers [...] constitutes a clear violation of creators' fundamental rights [...]."

Izyumenko does not dismiss the instinct behind such statements. "GenAI systems derive enormous economic value from vast amounts of human-created cultural material, so it seems only fair that the creators whose works make this possible should share in that value," she writes. The difficulty, according to the paper, is what happens next. "Copyright and related rights are often held and exercised by intermediaries (publishers, record labels, film studios) who acquire economic rights from original creators through assignments or licences. Strengthening copyright may therefore benefit those secondary copyright holders without necessarily improving remuneration of original creators."

That gives the paper its central question, in the author's words: "what does invoking creators' human rights actually require of an AI remuneration framework, beyond securing revenues for intermediary rightsholders?"

Two different rights wearing the same name

The legal core of the argument rests on a distinction that copyright debates tend to blur. Article 15(1)(c) of the International Covenant on Economic, Social and Cultural Rights (ICESCR), adopted on 16 December 1966 and in force since 3 January 1976, recognises the right of everyone "to benefit from the protection of the moral and material interests resulting from any scientific, literary or artistic production of which [she or] he is the author." Article 27(2) of the Universal Declaration of Human Rights contains a substantially identical guarantee.

Read expansively, those provisions could turn every copyright-derived payment into a human-rights entitlement, whoever holds it. The paper notes that versions of this reading have appeared in academic literature and in industry advocacy. One example it cites is a statement by the Canadian Copyright Institute to the World Intellectual Property Organization's copyright committee in November 2020, which argued that "[c]opyright is a human right, and strong copyright protections function in the public interest."

Izyumenko argues that this is hard to square with the UN Committee on Economic, Social and Cultural Rights (CESCR), the body of independent experts that monitors the covenant. Its General Comment No. 17, issued on 12 January 2006, describes intellectual property rights as legal entitlements whose scope, duration and transferability are set by law and can change. The human right, by contrast, protects the personal link between creators and their work and their basic material interests, grounded in human dignity and an adequate standard of living. General Comments are not legally binding, the paper acknowledges, though they carry considerable interpretative authority.

Who holds the right matters too. According to the paper, the CESCR clarifies that Article 15(1)(c) protects human creators, while intellectual property rights can be held by legal entities. European law does protect corporate copyright as property: the European Court of Human Rights has recognised that intellectual property held by companies can be a "possession" under Article 1 of Protocol No. 1 to the European Convention on Human Rights, and Article 17(2) of the EU Charter of Fundamental Rights states that "[i]ntellectual property shall be protected."

The paper treats that as a European peculiarity rather than a global rule. The right to property appears in Article 17 of the Universal Declaration but was not included as a freestanding right in either binding UN covenant. Even within Europe, the paper argues, the fundamental-rights status of property "does not make existing economic entitlements untouchable," pointing to the wide margin of appreciation the Strasbourg court gives states in regulating property, and to Martin Husovec's analysis that the Court of Justice of the EU has not recognised an inviolable core of intellectual property that would block legislative change.

From there, Izyumenko builds the paper's most consequential claim, and it is explicitly an argument rather than settled law. The material interests of original creators, she writes, "arguably warrant particular weight when balanced against the transferable economic entitlements of corporate copyright holders." The paper concludes that "there is therefore a strong human-rights argument for giving particular weight to creators' material interests when designing and regulating copyright, including through measures that limit or qualify the economic entitlements of corporate rightsholders." She is careful to add that this does not create a formal hierarchy in which creators automatically win.

The Commission's consultation in the crosshairs

Policy documents, the paper argues, routinely lose this distinction. It names the European Commission's call for evidence of 13 May 2026, which concerns a review of the Copyright in the Digital Single Market Directive and a targeted initiative for "a better copyright environment for European creativity and innovation." According to the paper, that document presents stronger protection for original creators and creative industries as a shared objective, and the subsequent targeted consultation groups their licensing and remuneration opportunities together "without addressing how licensing revenues would reach original creators."

The practical consequence is stated in one sentence: "an AI company could pay substantial fees to an intermediary without the authors and performers whose works are licensed receiving a corresponding share."

The paper presents this as an old problem with a new buyer. It cites the streaming economy, where, it says, authors and performers have not necessarily received a share corresponding to substantial revenue growth. It points to the European Commission's 2025 study of copyright contractual practices, which documents unequal bargaining power, buy-out agreements, low royalty rates and extensive transfers of exploitation rights, and to a study commissioned by the German Federal Ministry of Justice on remuneration in the streaming and platform economy.

How the EU framework works, and where the paper says it fails

The mechanics matter here. Article 4 of the Copyright in the Digital Single Market Directive (Directive 2019/790) creates an exception for reproductions and extractions of lawfully accessible works for text and data mining, unless the rightsholder has expressly reserved those rights. That reservation of rights is, in practice, an opt-out. The paper notes that the exception was not designed for generative AI but has since been understood to extend to AI training, a reading reflected in recital 105 and Article 53(1)(c) of the EU AI Act, which link general-purpose model training to the exception and to rights reservations under Article 4(3).

According to the paper, that produces two routes. Where rightsholders do not opt out, covered uses "require neither authorisation nor remuneration." Where they do opt out, further use needs another legal basis, possibly a licence, and whether money changes hands then depends on whether AI providers want the works, whether deals are signed, and on what terms.

Neither route, the paper argues, guarantees anything for the creator. The scale of AI training makes this worse than in streaming, according to Izyumenko. "Rather than negotiating with individual creators, AI developers have strong incentives to conclude agreements with publishers, producers and other intermediaries controlling large repertoires," she writes, citing her IViR colleague Martin Senftleben's observation that this may favour major repertoire owners while leaving smaller rightsholders and individual creators outside the licensing market altogether. Even a lucrative deal only reaches creators if their underlying contracts entitle them to a share of AI training revenue - which, the paper notes, they may not.

One point is easy to miss. The paper does not argue that the text and data mining exception is itself a human-rights violation. "Article 15(1)(c) ICESCR does not require payment for every individual use of a protected work," it states, and a footnote flags that this conclusion is the author's own interpretation of General Comment No. 17. The objection is structural. "Licensing opportunities, or even substantial licensing revenues, cannot therefore be taken as evidence that creators' material interests are adequately protected," the paper reads.

Transparency rules do not solve it either, in her view. The AI Act's copyright-compliance and training-content disclosure obligations under Article 53(1)(c) and (d) may help rightsholders find out which works were used and negotiate licences. "But knowing which works were used, and even securing payment for their use, is a different matter from ensuring that the resulting revenues reach their original creators," according to the paper.

Levies: money for culture, not necessarily for the trained-on

The second half of the paper tests the alternatives that academics and some policymakers have floated. They fall into two families, according to Izyumenko: output-based schemes, which would raise money from the commercial exploitation of generative AI systems, and input-based schemes, which would attach remuneration to the use of works in training.

Output-based levies would be triggered by the commercial offering or use of AI systems capable of substituting for human literary or artistic production. The money could go to creators or into social and cultural funds that finance artistic projects or help creators facing income losses. The paper credits these schemes with a real advantage: they avoid the need to identify every individual work used in training.

The human-rights problem is precision. General Comment No. 17 ties authors' material interests to their own productions. "Yet a fund supporting creativity generally might benefit creators whose works were never used for AI training, while providing no identifiable benefit to those whose works were," the paper states. In the LinkedIn post promoting the work, Izyumenko put the question more sharply, asking whether such a levy "could amount to a form of expropriation of their material interests under Article 15(1)(c) ICESCR."

The paper itself is more measured than the post. It says the difficulty "is not collective funding as such," and that collective administration of individual payments may well protect creators. The harder case is where collective funding replaces individual remuneration entirely. "In that case, protection of individual material interests is effectively replaced by support for the creative community more broadly," it reads.

Could human-rights law stretch to accept that? The paper raises the European Court of Human Rights Grand Chamber judgment in Verein KlimaSeniorinnen Schweiz and Others v Switzerland of 9 April 2024, where the court recognised the importance of collective mechanisms for effective protection in the climate context. Izyumenko notes that the judgment concerns collective standing, not the substitution of collective benefits for individual rights, and calls the question of a parallel under Article 15(1)(c) "an open question." Article 15(2) of the covenant, which requires measures for the conservation, development and diffusion of science and culture, may support collective measures, according to the paper, but does not displace individual protection. Her tentative resolution is a hybrid: an output-based scheme combining individual payments with cultural funding.

Individual rights: the problem of microscopic cheques

The input-based route comes in two forms. A statutory remuneration right would accompany an exception permitting training without permission, subject to payment. A residual remuneration right would keep exclusive rights and licensing in place but guarantee that original creators keep a claim to payment even after transferring their rights to an intermediary. The paper notes that residual rights already exist in EU law for rental under Article 5 of Directive 2006/115/EC and, in certain member states including Spain, Germany and Belgium, for online exploitation.

These models preserve a direct line between a creator's work and their payment, which is their appeal under Article 15(1)(c). Their weakness is arithmetic. "The vast number of works used in AI training could leave individual creators with payments too small to meaningfully protect their material interests," according to the paper. It asks directly: "If individual payments are merely microscopic, can they provide meaningful protection of those interests?"

Then it pulls back from its own question. The paper states that "there is not enough empirical evidence to establish how much creators would actually receive under different AI remuneration schemes," because amounts would depend on revenues collected, works covered and distribution rules. It also argues that the difficulty of identifying every work used in training does not necessarily make individual remuneration unworkable, citing proposals for metadata-assisted and probabilistic allocation with fallback distribution where precise matching fails.

The paper's most important qualification may be this: Article 15(1)(c) "does not necessarily require that remuneration for each particular use, or even for AI training as a whole, should be sufficient to provide creators with a livelihood." The test it proposes is narrower - whether creators "can meaningfully participate in the economic value generated through the use of their productions." Modest payments could meet that bar, it suggests, particularly where they accumulate or supplement other income.

What the paper does not claim

Several limits are stated in the text and are worth recording precisely.

The paper offers no model of its own. "Human rights law offers no 'magic' remuneration model," the abstract states, but instead "calls for examining whose interests different mechanisms protect, how effectively, and whether individual and collective approaches can complement one another."

It also sets aside freedom of expression. A footnote acknowledges that a full human-rights analysis of copyright and generative AI must consider the risk that copyright enforcement against AI outputs restricts lawful creative uses, but says that falls outside its focus.

Its conclusion is framed as a demand for scrutiny, not a verdict. "If creators' human rights are invoked to justify AI remuneration, we must therefore ask not only whether a proposed mechanism generates revenue, but who actually benefits from it, and whether the creators themselves are among those beneficiaries," the paper concludes.

Two disclosures appear on the first page. The ideas were first presented at an expert workshop titled "The Platform State Meets IP: Constitutional Accountability for Proprietary AI" at The City Law School in London on 8 June 2026, and refined at the USER Rights 2026 conference hosted by the Geneva Centre on Knowledge Governance on 1-2 October 2026. And, according to the paper, "OpenAI's ChatGPT was used to assist with language editing, improving clarity and structure, and refining the presentation of arguments," while the author developed the analysis, verified the legal sources and takes full responsibility for the content.

The administrative cost objection

The LinkedIn post drew 86 reactions and five comments by the time it was captured, and the replies previewed the practical objections such schemes will face.

Dennis W. K. Khong, who describes himself on LinkedIn as a law teacher with an interest in programming, technology and economics, raised the scale problem directly. "If 1 billion pieces of text went into training a text model, does it mean that all the copyright owners of that 1 billion text should be able to share the levy collected? It would appear the administrative cost of administering such a global levy system would out weight the amount collected," he wrote, adding that he was "all for collecting such a levy."

Linhua Emelie Lu, a law researcher at the University of Lorraine, asked whether competition law could address "the structural imbalances that prevent authors from receiving a fair share of the value generated by AI." Izyumenko replied by asking whether Lu had in mind "something like what's been done with competition law and press publishers' remuneration in France or Australia," calling it "definitely something worth looking into!" Elahe Avar, a legal and compliance assistant, summarised the paper's distinction: "fair remuneration is not just about whether creators are paid, but whether the payment genuinely protects their interests."

Why this matters for publishers and marketers

For PPC Land readers, the paper lands in the middle of a licensing market that has grown quickly while leaving the distribution question largely unanswered. Most of the deals and demands PPC Land has covered involve intermediaries on the selling side.

The largest US example remains Anthropic's settlement with authors. Anthropic agreed to pay at least $1.5 billion in September 2025 over roughly 500,000 works, about $3,000 per work in gross recovery - a figure that shows what per-work arithmetic looks like when a fixed pool is divided across a large catalogue. In the music sector, the Music Artists Coalition asked Universal Music Group in October 2025 for specifics on revenue splits from its Udio settlement, including whether payouts would simply offset artists' unrecouped balances - the precise scenario Izyumenko describes, in which money reaches the label but not necessarily the artist.

Collective licensing products are multiplying too. The Copyright Clearance Center now offers four AI licensing options, including pay-per-use transactional rights, though PPC Land's coverage noted that the announcement did not explain how payments reach individual authors. The US Copyright Office, in its May 2025 report on AI training, opposed compulsory licensing and floated extended collective licensing only where voluntary deals prove unworkable.

In Europe, the European Parliament resolution Izyumenko examines has already been cited by industry. When ARD, BDZV, MVFP, VAUNET and ZDF published a joint paper in April 2026, they drew on the same 10 March 2026 resolution to demand "appropriate remuneration" for media providers - a remuneration claim made by publishers and broadcasters, not by the journalists who write for them. That is not a criticism unique to the German coalition; the paper's point is that this framing is the norm. When 39 creative-sector organisations rejected the AI Act's implementation measures in July 2025, the signatories spanned collecting societies such as CISAC and GESAC alongside publisher bodies including the European Publishers Council and News Media Europe - the very mix of creators and intermediaries that the paper says policy too readily treats as one constituency.

Courts are adding pressure on the opt-out side. The Munich I Regional Court ruled against Suno on 31 July 2026 in a case brought by the collecting society GEMA, finding that the German text and data mining exception did not apply and setting fines of up to 250,000 euros per violation. In the UK, a House of Lords committee urged ministers in March 2026 to reject a broad commercial text and data mining exception, reporting that only 3% of consultation respondents favoured an exception with an opt-out.

The rights-reservation machinery underpinning all of this is technical. Under the EU's General-Purpose AI Code of Practice, finalised in July 2025, signatories commit to crawlers that follow robots.txt and to honouring other machine-readable reservations, with the AI Office overseeing compliance. For publishers that monetise through advertising, those opt-outs determine whether their pages feed training sets at all, and web scraping by AI crawlers has become a direct cost and a negotiating chip in the same breath.

The comment exchange about competition law also has a PPC Land precedent. The Court of Justice of the EU upheld Italy's rules requiring platforms to pay press publishers fair compensation in May 2026, in a case brought by Meta, with compensation tied to platforms' advertising revenue. That model, like most in this space, pays the publisher.

None of this means a statutory or residual right for creators is imminent. The Commission's review of the Copyright Directive is at the consultation stage, and the paper presents its analysis as a set of questions for that process rather than a draft. What it offers the creator economy and the publishers who rely on it is a sharper test for every AI licensing announcement: not how large the deal is, but whose bank account it ends in.

The author

According to her LinkedIn profile, Izyumenko has held her IViR post since September 2023. Before that she spent roughly a decade at the Center for International Intellectual Property Studies (CEIPI) in Strasbourg, from September 2012 to August 2023, where she completed her doctorate in December 2020. Her profile states that her work explores how copyright and trademark law in Europe and beyond intersect with the public interest and with human and fundamental rights. The paper cites her earlier work with Christophe Geiger, including the 2025 Edward Elgar commentary "Human Rights and Intellectual Property Before the European Courts," and a 2025 article with Senftleben on author remuneration in the streaming age.

Timeline

Summary

Who: Elena Izyumenko, Assistant Professor of Intellectual Property Law at the Institute for Information Law (IViR), University of Amsterdam. The paper concerns original creators (authors and performers), intermediary rightsholders such as publishers, labels and studios, AI developers, and EU policymakers.

What: A 15-page paper, "A Human Right to AI Remuneration? What Human Rights Law Tells Us Copyright Owes Creators," arguing that Article 15(1)(c) ICESCR protects creators' material interests rather than copyright ownership, that the EU's Article 4 text and data mining framework does not ensure AI revenues reach creators, and that both output-based levies and statutory or residual remuneration rights raise their own human-rights difficulties.

When: Written and posted on SSRN on October 16, 2026, and promoted on LinkedIn on October 17, 2026.

Where: Published on SSRN (id 7621139) from the University of Amsterdam; its analysis addresses EU copyright law, the EU AI Act, European human-rights law and the UN covenant.

Why: The European Commission is reviewing the Copyright Directive and consulting on AI licensing and remuneration, and the European Parliament has framed AI training compensation as a matter of creators' fundamental rights. The paper argues that such framing requires showing that payments actually reach creators, not only copyright owners - a question that bears directly on every AI licensing deal signed by publishers and labels.