How International Treaties Govern Copyright for AI-Generated Works
While domestic courts fracture over who owns AI outputs, the foundational treaties governing global copyright remain silent on whether an author must be human. This ambiguity leaves a massive legal gap for the international intellectual property system.
By Sofia Matos
- Strict Humanists
- Argue that 'intellectual creation' requires a human mind, meaning AI outputs belong in the public domain.
- Investment Protectors
- Argue that the effort of prompting and curating AI outputs deserves copyright to incentivize economic growth.
- Treaty Pragmatists
- Focus on the need to update international frameworks to prevent a collapse of cross-border copyright recognition.
Perspectives this story doesn't cover
- Independent artists whose works were used in training data
- Open-source AI developers
Summary
- International copyright treaties like the Berne Convention do not explicitly define an 'author' as a human being.
- The WIPO Copyright Treaty protects the expression of ideas, raising questions about whether an AI prompt is an idea or an expression.
- The EU generally requires human personality for copyright, while China has begun protecting AI outputs based on human investment.
- This divergence threatens the core purpose of international treaties: the mutual recognition of copyright across borders.
- The standard copyright duration of 'life plus 50 years' cannot be applied to non-biological AI systems.
One faction of legal scholars reads the foundational texts of global intellectual property and sees a deliberate silence: because the Berne Convention never explicitly uses the word "human," they argue member states are perfectly free to assign copyright to the algorithms that generate art, or to the companies that deploy them. An opposing faction reads those exact same treaties and sees a biological absolute: the phrase "intellectual creation" inherently demands a human mind, meaning any image or text generated by a machine belongs instantly to the public domain. Both sides are currently advising national governments on how to rewrite their intellectual property laws.[3][5][7]
This is not merely an academic debate over semantics. As generative models produce billions of images, lines of code, and synthetic texts daily, the financial stakes of who owns them are reshaping the global economy. Yet the rules governing this ownership are not being drafted by modern technologists; they are being retrofitted onto international agreements drafted decades, or even centuries, before neural networks existed.[7]
To understand how the world is handling AI authorship, it is necessary to look past domestic rulings and examine the international treaties that bind 181 countries together. The most important of these are the Berne Convention of 1886 and the WIPO Copyright Treaty (WCT) of 1996. These documents establish the baseline for how copyright is recognized across borders, ensuring that a novel written in France is protected from piracy in Japan.[1][7]
The WCT was specifically designed to bring copyright into the digital age, addressing the protection of computer programs and databases. However, it focuses entirely on the nature of the work rather than the biological nature of the creator. As Article 2 of the WCT explicitly states: "Copyright protection extends to expressions and not to ideas, procedures, methods of operation or mathematical concepts as such."[1]
Because the treaty protects the "expression," the critical legal question for generative AI becomes whether a prompt typed into a model constitutes the expression itself, or merely the idea. If the prompt is just an idea, the AI is doing the expressing, and the human operator has no legal claim to the output under the WCT framework.[1][6]
This ambiguity has led to a fractured global landscape. In the European Union, the legal framework relies heavily on the standard of the "author's own intellectual creation." European courts have historically interpreted this standard as requiring human personality and free, creative choices to be imprinted upon the final work.[4]
Under this strict EU interpretation, if an AI model makes the final choices about the composition of an image, the color palette, or the phrasing of a paragraph, the human prompter lacks sufficient creative control to claim authorship. The machine acts as a black box, severing the link between the human's intent and the final expression.[4][5]
The machine acts as a black box, severing the link between the human's intent and the final expression.
Conversely, Chinese courts and policymakers are testing a radically different interpretation of these same international boundaries. Recent rulings in China have suggested that the intellectual investment required to craft specific, iterative prompts can satisfy the threshold for originality, treating the AI as a highly advanced tool rather than an autonomous creator.[2]
By recognizing the human prompter's effort as the creative act, Chinese policy effectively grants copyright protection to AI-generated works without violating the Berne Convention's silence on the matter. This approach prioritizes economic incentive, ensuring that companies investing heavily in AI generation can protect their digital assets.[2][7]
Legal scholars note that this divergence is possible exactly because the WIPO treaties do not define the word "author." When the treaties were drafted in 1886 and updated in 1996, the assumption that an author was a human being was so universally understood that it did not need to be codified into international law.[3][5]
Now, that unwritten assumption is the central vulnerability of international copyright law. If a synthetic image is protected as a valuable commodity in China but considered uncopyrightable public domain material in the EU, the core purpose of the Berne Convention—mutual recognition of copyright across borders—begins to break down.[2][4][7]
The World Intellectual Property Organization (WIPO) has recognized this impending crisis. In ongoing policy discussions, WIPO has formally asked member states to consider a foundational question: "Should copyright be attributed to original literary and artistic works that are autonomously generated by AI or should a human creator be required?"[6]
If the international consensus decides that a human is strictly required, the secondary question becomes how much human input is "enough" to qualify. Is selecting the training data a creative act? Is writing a 500-word prompt? Is curating and editing the final output? The treaties currently offer no metric for measuring human contribution.[6][7]
Furthermore, the Berne Convention mandates a minimum copyright term of the author's life plus 50 years. If an AI is deemed the author, it has no biological life and no date of death, making the standard duration metric impossible to apply without sweeping legislative reform.[3][7]
To bypass this biological hurdle, some legal theorists propose creating a new "neighboring right" for AI-generated works. Similar to the rights granted to broadcasters or phonogram producers, this would protect the economic investment of generating the work without granting full traditional authorship or requiring a human lifespan.[5]
Until a new international consensus is reached, creators and technology companies must navigate a patchwork of national laws. A digital asset might be a protected, monetizable commodity in one jurisdiction and freely copyable in another, fundamentally altering how global media franchises and software developers operate.[7]
The resolution of this gap will not come from a single domestic court case, but from how the 181 member states of the WIPO Copyright Treaty choose to interpret the silence of their predecessors. The definition of human creativity is no longer just a philosophical question; it is the most pressing treaty negotiation of the digital era.[1][7]
Definitions
- Berne Convention
- An 1886 international agreement governing copyright, which established the principle that copyrights are automatically recognized across member states.
- WIPO Copyright Treaty (WCT)
- A 1996 international treaty that updated copyright laws for the digital environment, specifically protecting computer programs and databases.
- Public Domain
- Works whose intellectual property rights have expired, been forfeited, or are inapplicable, making them free for anyone to use.
- Neighboring Rights
- Legal rights related to copyright that protect the financial and organizational investment in a work, such as the rights of broadcasters or record producers.
Questions & answers
Do international treaties explicitly ban AI copyright?
No. Treaties like the Berne Convention and the WIPO Copyright Treaty do not explicitly define the word 'author,' leaving a gap regarding non-human creators.
How does the EU handle AI authorship?
The EU relies on the standard of the 'author's own intellectual creation,' which courts generally interpret as requiring human personality and free choices, making AI copyright difficult to obtain.
Can a text prompt be copyrighted?
It depends on the jurisdiction. Some countries view prompts as mere ideas, which are not protected, while others see highly detailed, iterative prompts as a form of creative expression.
Significance
The rules determining whether a company can own the code, text, or art its AI generates are not being written from scratch; they are being retrofitted onto 19th-century international agreements. How member states interpret these treaties will dictate the financial structure of the global generative AI economy.
Sources
[1]WIPOTreaty PragmatistsWIPO Copyright Treaty (WCT)
Read on WIPO →
[2]Michigan Journal of International LawInvestment ProtectorsHow Human is Human Enough: The Implications of Chinese Policy on AI Generated Copyright Protection
Read on Michigan Journal of International Law →
[3]Wolters KluwerTreaty PragmatistsOriginality in the Age of AI: How to Get Copyright Protection Now?
Read on Wolters Kluwer →
[4]Cairn.infoStrict HumanistsThe future of AI and authorship: the EU's legal framework for AI and copyright
Read on Cairn.info →
[5]DigiConTreaty PragmatistsAI Authorship and Copyright Protection
Read on DigiCon →
[6]WIPOTreaty Pragmatists(i) Should copyright be attributed to original literary and artistic works that are autonomously generated by AI or should a human creator be required?
Read on WIPO →
[7]Factlen Editorial TeamTreaty PragmatistsSynthesis by Factlen editorial team
Read on Factlen Editorial Team →
Comments
More in Artificial Intelligence
See all →Reinforcement Learning
The Mathematical Tuple That Governs Autonomous AI Decision-Making
8 sources
AI Hallucinations
The Three Primary Causes of Hallucination in Large Language Models
6 sources
AI Accountability
FBI and DOJ Weigh Criminal Liability for Autonomous AI After Agents Access US Government Sites
3 sources
AI Coding Impact
The Impact of AI Automation on Software Engineering Jobs
2 sources
Every angle. Every day.
Get Artificial Intelligence stories with full source coverage and perspective breakdowns delivered to your inbox.




