From specificatio to generative AI: should EU Law recognise meaningful human transformation?

Efi Thoma, Senior Global Legal Counsel at TeKnowledge in Cyprus, draws on an ancient property-law principle to question whether EU law should recognise meaningful human transformation of AI-generated content.

Generative AI is testing a familiar assumption of intellectual property law: when a person uses technology to produce something new, at what point does their contribution become legally significant?

A generic instruction to an AI system, followed by acceptance of its output, provides limited evidence that the resulting expression reflects the user’s own intellectual creation. Conversely, where AI functions as a tool within a genuinely human creative process, existing copyright law may already provide protection.

The more difficult question lies between these two situations. What happens when a user provides detailed instructions, evaluates and rejects outputs, repeatedly refines the material, makes deliberate selections, restructures or edits the result, and meaningfully influences the final output, yet their contribution does not satisfy the copyright threshold?

Current EU law may leave some commercially valuable, human-directed AI outputs without a general harmonised exclusive right. That does not necessarily reveal a regulatory gap. Leaving material outside exclusive protection may itself promote competition, innovation and access to information. It does, however, raise an important policy question: should meaningful human transformation of AI-generated material sometimes attract a limited legal consequence even where copyright does not?

An unexpected source of inspiration may lie in the ancient property-law doctrine of specificatio.

Greek law recognises ειδοποιία, best translated in comparative legal terminology as “specification” (specificatio), under Article 1061 of the Greek Civil Code. Broadly, where a person produces a new movable thing by processing or transforming material belonging to another, ownership may arise in the processor where the statutory conditions are satisfied, including the requirement that the value of the work be significant. Significant transformative contribution can justify a new legal consequence. That principle may offer useful inspiration for AI regulation.

The relevance of specification to generative AI should not be overstated. This legal doctrine does not apply to AI-generated outputs. The connection is conceptual rather than doctrinal. It lies in the idea of legally significant transformation. Specificatio reflects a longstanding legal willingness to consider whether a sufficiently significant transformative contribution can justify a new legal consequence. That principle may offer useful inspiration for AI regulation.

This leads to a possible new legislative concept: Digital Specification.

Digital Specification could describe the lawful production or transformation of a generative-AI output through a meaningful, causally demonstrable and qualitatively significant human contribution where the relevant output does not otherwise qualify for copyright protection. The corresponding Digital Specification Right would be a proposed new statutory concept – not an interpretation of existing EU copyright or property law. It should not amount to traditional ownership of information, but rather to a narrowly defined sui generis entitlement against particular forms of commercial appropriation.

Human contribution would be central, but it should not be reduced to the mechanical act of prompting. EU law has not established a copyright test based on the number, length or sophistication of prompts. The legally relevant question remains whether identifiable human creative choices are expressed in the resulting material. Accordingly, generic or minimal prompting would ordinarily provide weak evidence of copyrightable human authorship, but it should not be treated as categorically incapable of contributing to copyright protection.

Equally, extensive prompting, repeated iterations or substantial time spent using an AI system should not automatically establish authorship. Effort is not a substitute for originality. The same caution is necessary when considering human validation and critical thinking.

Reviewing an AI-generated result, checking its accuracy or approving it for use should not, by themselves, create an intellectual property entitlement. Validation is not equivalent to authorship, and professional expertise does not automatically convert machine-generated expression into human-created expression.

Critical judgment may nevertheless become relevant where it forms part of a broader transformative process. If a user identifies deficiencies, rejects or corrects generated material, applies contextual or professional judgment, and uses that assessment to materially shape subsequent iterations or the final result, those actions may provide evidence of meaningful human involvement.

The distinction is important: the legally relevant factor would not be validation itself, but whether human judgment causally influenced the form, structure, content or function of the resulting output.

This may become increasingly significant in professional uses of generative AI. A lawyer, researcher, engineer or other professional may add considerable value by identifying errors, testing assumptions, determining relevance and deciding whether an AI-generated result can responsibly be used. Those activities may be essential from a professional responsibility, governance or risk-management perspective. They should not, however, automatically be equated with copyright authorship.

Where the user’s contribution produces identifiable expression reflecting their own intellectual creation, ordinary copyright may apply to that expression. Importantly, an AI-assisted result need not be classified as wholly copyrightable or wholly unprotected. Human-authored elements may coexist with AI-generated, factual, public-domain or otherwise unprotected material.

Digital Specification would therefore address only a narrower intermediate category: meaningful and demonstrable human transformation that does not satisfy existing copyright requirements.

Even there, the proposed right should not become “copyright-lite.” Creating substantially the same exclusive rights as copyright while applying a lower originality threshold would risk undermining copyright’s deliberate boundaries. A more defensible model would target only defined forms of knowing and substantially substitutive commercial appropriation.

The right should not prevent independent creation or independent AI generation. Nor should it confer exclusivity over ideas, facts, data, methods, styles or material already free of copyright protection. Legitimate quotation, criticism, research, teaching and genuinely transformative reuse should remain available. Any protection should also be deliberately short, with its duration determined through economic and empirical assessment rather than derived from specificatio itself.

Equally important, Digital Specification must never become a mechanism for converting infringement into a new entitlement. Existing copyright, database rights, trade secrets, confidentiality obligations, data-protection requirements and other third-party rights would remain unaffected. Processing protected material through an AI system should not transform unlawful use into a new proprietary claim.

This distinction is particularly relevant in the European context. The EU AI Act, Regulation (EU) 2024/1689, as subsequently amended, now contains significant obligations concerning AI governance, transparency and copyright compliance. Article 53 requires providers of general-purpose AI models, among other things, to maintain a policy to comply with EU copyright law and to publish a sufficiently detailed summary of the content used for training. Article 50 establishes differentiated transparency and marking obligations for certain AI-generated or manipulated content.

These provisions are important, but they do not determine copyright authorship or ownership of AI-generated outputs, nor do they create a new exclusive right in such outputs.

EU copyright law likewise contains no specific binding rule determining the copyrightability, authorship or ownership of generative-AI outputs. Existing copyright principles therefore remain the starting point. Database rights, trade-secret protection, contract, designs and other legal regimes may protect particular interests where their respective requirements are satisfied, but none currently creates a general harmonised entitlement merely because a user meaningfully directed or refined an AI-generated result.

The European Parliament’s March 2026 resolution on copyright and generative AI demonstrates that these questions are firmly on the EU policy agenda, but the resolution is non-binding. Further EU work on AI and copyright is continuing, including Commission consideration of licensing, remuneration and enforcement issues. As of October 2026, however, no EU legislation has created a general sui generis right specifically for AI-generated outputs.

Whether such a right should exist remains an open policy question.

The objections are substantial. A new entitlement could increase transaction costs, encourage claims over commonplace AI outputs, generate difficult evidentiary disputes over prompts and iterations, restrict competition and unnecessarily remove material from the sphere of lawful reuse. The fact that an AI-generated result has commercial value does not, by itself, justify granting someone an exclusive right over it.

But if future evidence demonstrates a genuine problem of commercial appropriation of substantial human-directed AI work falling outside copyright, expanding copyright may not be the most coherent response. A narrower legal interest focused specifically on meaningful human transformation could merit consideration.

The resulting architecture would be deliberately restrained. Human intellectual creation would remain within copyright. Predominantly autonomous AI generation would receive no new exclusive protection under the proposed framework. Between those categories, EU lawmakers could consider whether demonstrable human transformation- potentially evidenced by purposeful direction, critical judgment, refinement and validation that materially shapes the result- justifies a limited sui generis right against narrowly defined commercial appropriation.

Roman lawyers debating specificatio could hardly have anticipated generative AI. Yet the underlying legal question is surprisingly familiar: when existing resources and human intervention combine to produce something new, when does the transformative contribution become sufficiently significant for the law to recognise it?

The answer is not that specificatio applies to AI. It does not. The more interesting proposition is that its underlying principle of legally significant transformation may provide a useful conceptual starting point for future regulation.

Perhaps, then, the next question for AI law should not simply be “Who owns AI-generated content?”

It should be:

When has human judgment moved beyond merely requesting or validating an AI-generated result and become sufficiently transformative for the law to consider recognising a limited legal interest in the resulting output?

That distinction, between human oversight and human transformation, may become increasingly important as generative AI becomes embedded in professional and creative work.

This article reflects the author’s personal views and proposes “Digital Specification” as a concept for legal and policy discussion.

Related Publications