EU flag beside books, creative artwork and an AI network screen in a policy office. Filename: eu-ai-copyright-consultation-closes-3-november.webp

EU AI Copyright Consultation Closes Today: What Brussels Is Asking

ByAIERA Blog Editorial Team

Key Takeaways

  • The EU AI copyright consultation closes on 3 November 2026, after opening on 29 September 2026, and it is a targeted consultation, not a new law: nothing in it changes any obligation today.
  • The consultation covers four areas: the use of copyright-protected content in AI systems, online piracy of time-sensitive content such as live events, a single equitable remuneration right for music performers and producers, and copyright in scientific research.
  • Its most watched question is AI training data: the Commission is asking how the existing copyright framework handles the use of protected works to train generative AI, including licensing, enforcement, and the current text-and-data-mining exception with its opt-out.
  • A European Parliament resolution adopted on 10 March 2026 pushed in a more radical direction, proposing a shift from opt-out to opt-in for training data, but a Parliament resolution is not law, and the consultation itself proposes no legislation.
  • What happens next is assessment, not enactment: the Commission will weigh the responses before deciding whether any further measures are warranted, and coverage of the process expects a formal proposal, if one comes at all, no earlier than 2027.

Introduction

The EU AI copyright consultation reaches its deadline today, 3 November 2026, and it is worth being precise about what that deadline is — and what it is not. It is not the day a new European AI copyright law takes effect. It is not the day AI companies must change how they train models. It is the last day for stakeholders to answer a set of questions the European Commission put to them on 29 September 2026 about whether the existing copyright framework still works when generative AI is trained on, and generates, vast amounts of protected content.

That distinction matters because the consultation sits at the intersection of two different pressures. Rights holders argue that their work is being used at scale without permission or payment. AI developers argue that training is already governed by existing exceptions, licensing markets, and opt-out mechanisms, and that changing the rules now could slow European AI development. The consultation does not resolve that argument. It gathers evidence and views that could inform future measures — binding legislation, voluntary cooperation, or non-legislative tools — and the Commission has said it will assess that evidence before deciding whether further action is warranted.

For anyone who publishes, licenses, builds with, or simply uses generative AI in Europe, the consultation is therefore a map of the questions Brussels is actually asking, rather than a preview of answers. This article walks through the four areas, the training-data question at the centre, the opt-in proposal that is — and is not — on the table, and what to watch after the window closes.

EU AI copyright consultation: the four questions

The consultation is described by the Commission as a targeted consultation on the challenges posed by technological developments, including artificial intelligence, for copyright protection, and on options for potential future measures. It builds on a call for evidence conducted earlier in 2026, along with evidence gathered in previous studies and assessments. The invite list is deliberately broad: copyright and related-rights holders, generative AI providers and other actors in the AI value chain, intermediaries, collective management organisations, sports and other live-event organisers, research organisations, national authorities, consumer organisations, and non-governmental organisations.

The four areas are:

  1. The use of copyright-protected content in AI systems. This is the training-data question, and it dominates the coverage. Dedicated sections examine AI licensing and enforcement, AI-generated imitations of performers, and the use in the EU of sound recordings of non-EU nationals, according to legal summaries of the consultation.
  2. Online piracy of content, including live events. Here the issue is time: a live broadcast or event loses most of its value if pirated copies circulate while it is still happening, so the consultation asks about enforcement against piracy of time-sensitive content.
  3. The single equitable remuneration right of music performers and producers. This asks whether performers and producers should receive a single, equitable payment when their recordings are used, and how such a right would work in practice.
  4. Copyright and scientific research. This covers how researchers can use protected material, including in AI-assisted research, without the rules becoming either a free-for-all or a paperwork trap.

None of these four areas arrives with a draft law attached. Legal summaries are explicit on this point: at this stage the consultation does not introduce a legislative proposal or change existing obligations. That is the single most important sentence for readers who might otherwise assume that 3 November brings new rules.

The training-data question at the centre

To understand why the training question is so contested, it helps to start with the current EU position as described in coverage of the consultation. EU copyright law includes a text-and-data-mining exception that allows AI training unless rights holders explicitly opt out in a machine-readable format. In practice, that means a publisher or creator who does not want their work used for training must signal that refusal — for example through robots.txt or a similar machine-readable signal — and the default, absent such a signal, is that mining is permitted under the exception.

That opt-out model is exactly what the recent shared opt-out signal debate is about. On 30 October 2026, OpenAI, Google, Microsoft, and Anthropic said they would back one shared robots.txt token for publishers who want to opt out of AI training, replacing a patchwork of per-company crawler instructions with a single signal. That announcement is separate from the Commission consultation, but the two are clearly related: the consultation asks whether the legal framework behind such signals is adequate, while the companies’ announcement is a voluntary, technical step within the existing framework. One does not replace the other, and neither, on its own, settles the legal question.

The consultation asks, in effect, three things about training: whether the current exception and opt-out work in practice, whether licensing markets are functioning well enough that rights holders can actually get paid, and whether enforcement is realistic when a model has already been trained on disputed material. These are evidence questions, not rhetorical ones. A rights holder who can show that opt-outs are ignored, or that licensing offers are never made, is giving the Commission a different kind of evidence than a developer who can show that licensing markets are growing and that opt-outs are honoured. The responses, not the questions, will shape what follows.

The opt-in proposal: real, but not the Commission’s proposal

One idea that will appear in many responses, and in much of the commentary, deserves careful handling: the shift from opt-out to opt-in. On 10 March 2026, the European Parliament adopted a resolution on copyright and generative AI proposing a more radical direction — shifting the training-data authorisation mechanism from opt-out to opt-in, so that AI developers would need explicit permission before using protected works.

That resolution is real, and it is a significant political signal. It is not, however, a Commission proposal, and it is not part of the consultation document itself. A Parliament resolution expresses a political position; it does not amend the Copyright Directive or create a new obligation. Treating it as though Brussels had already decided on opt-in would misread the process. What the resolution does is raise the stakes of the consultation: if the Commission’s evidence-gathering shows that the opt-out model is failing in practice, the opt-in idea is already on the table as an alternative, with parliamentary backing behind it. If the evidence shows the current model working, the resolution remains a position, not a law.

The same caution applies to a US case that has been cited alongside the consultation. In the dispute between Thomson Reuters and Ross Intelligence, a US Court of Appeals rejected a fair-use defence for using Westlaw headnotes to train a competing AI legal-research product. That is a US decision under US fair-use law, not an EU ruling, and it does not change EU rules. It is relevant as context — it shows courts scrutinising training-data claims in other jurisdictions — but it is not evidence of what the Commission will do.

What the consultation is not

It is easy, in a fast-moving policy debate, to overstate a consultation. Three common misreadings are worth correcting explicitly.

First, the consultation is not an enforcement action. The Commission is not, through this document, fining a company, ordering a model to be retrained, or declaring a particular training practice unlawful. Enforcement of the general-purpose AI rules under the EU AI Act began on 2 August 2026, under a separate track, and the consultation does not extend or change that enforcement.

Second, the consultation is not limited to generative AI chatbots. Its four areas deliberately reach beyond model training into live-event piracy, music remuneration, and research — areas where AI is one factor among several, and where the affected industries have their own, older disputes with the copyright framework. A response that treats the whole exercise as “the AI training consultation” misses three quarters of it.

Third, the consultation does not promise a law. The Commission’s own framing is that it is gathering feedback on challenges and options for potential future measures. “Potential” is doing real work in that sentence. The outcome could be a legislative proposal, a voluntary-cooperation initiative, a non-legislative measure such as guidance or standard-setting support, or — if the evidence does not support action — no new measure at all. Coverage of the process notes that, based on the usual sequence of impact assessment followed by draft legislation, a formal proposal is unlikely to take shape before 2027 at the earliest. That is an expectation drawn from EU legislative practice, not a date the Commission has announced.

What to watch after 3 November

Once the window closes, the process moves out of public view for a period. The Commission will assess the responses, and the next public signal will be whatever it chooses to publish about that assessment — which could be a summary of responses, an impact assessment, or a proposal, on a timetable it has not fixed. Three things are worth watching in the meantime.

The first is whether the Commission signals a preferred direction early, for example by publishing a summary that emphasises licensing gaps or, conversely, the functioning of existing markets. The second is how the AI training opt-out signal announced on 30 October performs in practice: if publishers adopt it widely and companies honour it, the opt-out model’s defenders gain evidence; if adoption is low or honours are patchy, its critics gain evidence. The third is the interaction with the AI Act’s general-purpose AI track, where enforcement powers are already active: copyright questions and AI safety and transparency questions are travelling on parallel tracks, and a development on one track will inevitably be read as a signal about the other, even when the legal basis is different.

For publishers and creators, the practical step before any of that is unglamorous but concrete: know what your current opt-out and licensing position actually is, and whether it is machine-readable and up to date. For AI developers and deployers, the equivalent step is to be able to show, with records, what data was used, what licences were obtained, and what opt-outs were honoured. Whatever the Commission decides, that kind of record is the evidence both sides will need — and it is the kind of record that is much harder to reconstruct afterwards than to keep now.

The consultation closes today. The debate it is gathering evidence about will not. If Brussels does legislate on AI copyright, the shape of that legislation will owe a great deal to the responses filed by 3 November — which is why the deadline matters, even though no law changes today.

For a related explainer on how AI systems are used in practice, see AI Agents Explained: What They Are, How They Work & Uses, and for the tools shaping this market, see Best AI Tools 2026: What’s Worth Paying For and What to Skip.

Sources

Similar Posts