AI is built on a centuries-old human cultural commons but its dividend flows to a few US firms. Can the EU capture a public share?

► President
I open this on a provocation that deserves a serious answer. The argument: AI is built on centuries of human culture and knowledge — a shared commons — yet the economic dividend of that pooling flows almost entirely to three private US firms and their shareholders. Europe can regulate as much as it likes, but if it remains a spectator, regulation does not change who captures the value. The question has three layers. The moral and economic claim: if frontier AI is trained on the collective output of human civilisation, including European literature, science and art, is it just that the returns accrue privately and extraterritorially? The legal question: does the EU have instruments to capture a share — through copyright and the text-and-data-mining regime, through taxation, through a public stake, or through building public AI capacity? And the strategic question, the tweet's sharpest point: regulation without capability is spectator regulation. The legal anchors are the DSM Copyright Directive and its text-and-data-mining exception, the AI Act transparency obligations on training data, Article 114 TFEU for any internal-market measure, and the unanimity-bound tax route for any levy. Three tensions: copyright remuneration versus innovation access; value capture versus competitiveness; and regulatory power versus the absence of European AI capacity.
[EUR-Lex — verify DSM Directive 2019/790 Arts. 3-4; AI Act GPAI transparency; TFEU Arts. 113, 114, 311]

► EVP · Economy
This portfolio leads, with Justice on copyright. The claim is partly right and partly a category error. Right: the training input is in significant part a commons — human-generated text, images, code, scientific literature — and the returns to frontier models are concentrated in a few non-EU firms. The category error: AI value is not only captured from the commons; it also reflects enormous private capital in compute, talent and engineering. So the question is not who owns the commons, which by definition no one does, but whether the EU can capture a fair share of the value that European inputs and the European market generate. Four instruments, in ascending legal difficulty. One, copyright remuneration — the most actionable: under the DSM Directive rightsholders can opt out of the text-and-data-mining exception, and the AI Act requires GPAI providers to disclose training-data summaries and respect opt-outs, creating a licensing market that channels value to European creators. Two, a public AI fund or sovereign stake via the EIB and InvestEU — feasible but fiscally large. Three, an AI levy or data dividend — a tax measure requiring unanimity under Articles 113 and 311, politically near-impossible and risking offshoring. Four, public AI capacity — the EuroHPC AI factories and open European foundation models, the only instrument that addresses the spectator critique directly. My recommendation: a strong copyright-remuneration market plus public AI capacity. Reject the levy as unanimity-blocked and counterproductive. The tweet's deepest point stands — capture requires capability.
[EUR-Lex — verify DSM Directive Arts. 3-4; AI Act GPAI; TFEU Arts. 113, 311]

► Commissioner · Justice
Co-lead, on the copyright frame — the most legally grounded route. The DSM Directive's text-and-data-mining exception already lets rightsholders reserve their rights, opting out of commercial mining. The AI Act reinforces this by requiring GPAI providers to put in place a policy to respect copyright and to publish a training-data summary. The combination creates the licensing market the commons argument demands: European publishers, authors, musicians and image-rights holders can require payment. My condition: the opt-out and transparency obligations must be enforced — today they are widely ignored, and an unenforced right captures nothing. I also flag that the commons in the strict sense — public-domain works, facts, ideas — is not copyrightable, so remuneration captures the protected layer, not the genuinely public-domain layer. The public-domain layer is best served by public AI capacity.
[CJEU — verify TDM exception scope; DSM Directive Arts. 3-4]

► EVP · Digital
Support, with the capability point sharpened. The tweet is right that regulation without capacity is spectating. The AI Act regulates how AI is built and deployed; it does not give Europe its own frontier models. The answer to value capture is the EuroHPC AI factories plus open European foundation models — public infrastructure on which European firms and the public sector build. The training-data transparency obligation in the AI Act is also the enabler of copyright remuneration: you cannot license what you cannot see, so enforce it robustly. But I caution against a levy — it would push the few European AI efforts offshore and entrench the US lead.

► Commissioner · Internal Market
Support for public capacity, reservation on remuneration design. The copyright-licensing market is right in principle but must not become a compliance thicket that only the US giants can navigate — they can afford licensing deals, European startups cannot. Design the remuneration system so it does not entrench the incumbents it is meant to discipline. On capacity, the EuroHPC AI factories and a European open-model strategy are the real lever; pair them with the Chips Act for the compute layer.

► Commissioner · Competition
Caution, legally precise. Two warnings. First, a value-capture mechanism targeting only non-EU firms is discriminatory and breaches national-treatment rules — Trade will confirm. Second, collective licensing of training data must not become a cartel of rightsholders fixing prices; it must be structured within the competition rules, which is why collective-management organisations are regulated. I support copyright remuneration and public capacity, but the design must be competition-compliant.

► Commissioner · Trade
Concurring with Competition. Any levy or remuneration scheme must apply equally to EU and non-EU AI providers. A measure designed to extract value specifically from US firms would trigger a trade dispute and likely retaliation — we have seen this with digital-services taxes. The copyright-licensing market is trade-defensible because it applies to all who train on protected works, regardless of origin.

► Commissioner · Budget
The own-resources angle. If the College ever wanted an AI dividend to flow to the EU budget rather than just to private rightsholders, that would be a new own resource — requiring unanimity and national-parliament ratification under Article 311. Realistically out of reach this mandate. The more feasible public-benefit route is funding EuroHPC and open European models from existing programmes, so the public captures value through capability, not through a new tax.

► Commissioner · Education & Culture
Strong support for the cultural-commons recognition. European cultural heritage — literature, art, film, music — is a core training input, and European creators deserve remuneration when their work trains commercial AI. The opt-out and AI Act transparency are the tools. And public European models should be trained on European cultural and linguistic data so European languages and cultures are well-served, not flattened by anglophone models. Cultural sovereignty is part of the dividend.

► Commissioner · Research & Innovation
Support for public capacity. European research generates much of the foundational knowledge; open European models and public compute let that knowledge return public value rather than only feeding private frontier labs. Fund the EuroHPC AI factories and European open-model consortia.

► President
I read the room. Clear majority that the value-capture question is legitimate and currently unaddressed, and that the answer is copyright remuneration plus public AI capacity, not a levy. The tensions: capture by rule versus capture by tax — Economy, Justice and Digital converge on enforcing the copyright opt-out and AI Act transparency, channelling value to European creators, plus building public AI capacity that returns value to the public; Budget and Competition are firm that a levy or new own resource is unanimity-blocked and trade-risky. Remuneration versus innovation access — Internal Market and Research warn that a licensing market must not become a thicket only incumbents can navigate, so design for SME access. The spectator critique — Digital and Economy concede the tweet's deepest point: regulation without capability is spectating, and the only instrument that truly answers value capture is European public AI capacity, which connects to the digital-sovereignty debate. The non-discrimination guardrail — Competition and Trade: any mechanism must apply equally to EU and non-EU firms or it breaches trade law and invites retaliation. No blocking objection; the levy is rejected on legal and economic grounds, not blocked by a portfolio. I record consensus: the EU can capture a fair share through an enforced copyright-remuneration market and public AI capacity, but not through a discriminatory levy or a new own resource. Regulation alone is spectating; the answer is capability. No formal vote.
[review — political judgement required on the appetite for public AI investment at scale]

Deliberation closed · 11 positions recorded