AI and the Future of Art: Rethinking Protection Beyond Human Authorship

“AI art isn’t real art.” That claim followed Théâtre D’opéra Spatial, the image Jason Allen created using Midjourney that won first prize at the 2022 Colorado State Fair Fine Arts Competition. Allen spent over 80 hours refining prompts and editing the final image, but many dismissed the work as machine-made. The controversy reignited an old question: do we value art for the creativity that produces it or for the form that meets the eye? That same question now confronts copyright law. 

If a human shapes the prompt, why should AI tools be treated any differently than a pen or a brush? AI-generated art is already testing the limits of copyright, and to understand how the law should respond, we need to look at what happens before the prompt, after the prompt, and in the legal grey zones in between. 

Before the prompt: 

Before a user writes a single prompt, developers train AI models on vast datasets of human-made images, often collected without permission and used without compensation. This practice raises a fundamental conflict, where the freedom to innovate collides with the rights of artists to control the use of their work. Three legal issues follow: unauthorized use of copyrighted material, economic harm caused by AI-driven mimicry, and the uncertain reach of existing exceptions under the Copyright Act. 

Developers train AI tools like Stable Diffusion, Midjourney, and DALL-E 2 on billions of scraped images created by human artists. Drawing on those datasets, these systems can reproduce the look and feel of a specific artist with striking accuracy. In some cases, they even replicate recognizable elements of that artist’s work, as seen in trends like recreating photos “in the style of Studio Ghibli.” Through this process, AI starts to feel less like a neutral tool and more like a digital mimic. 

AI’s ability to mimic artistic style threatens to take a direct economic toll on artists. Canada’s creative industries generate around $60 billion a year. Now, artists must compete with tools that can produce content instantly, at scale, and without compensation. Even if AI systems draw inspiration from copyrighted works, the law protects expression, not style; colour palettes, brushwork, and general “feel” are not protectable. That leaves artists exposed to what feels like appropriation but technically is not infringement. The backlash to Théâtre D’opéra Spatial made this tension visible. Jason Allen’s win at the Colorado State Fair became a focal point not because the image lacked effort or beauty, but because its style felt borrowed and its creation, in some eyes, unearned. 

Even if artists could craft an infringement claim against AI developers, copyright’s exceptions may shield those developers from liability. Section 30.71 of the Copyright Act permits temporary reproductions, but only when they are part of a technical process, exist briefly, and support a non-infringing use. Meanwhile, section 29 allows fair dealing for research, a category that courts have interpreted with considerable flexibility, including in some commercial contexts. These provisions give developers room to argue that training AI on copyrighted material is lawful. At the same time, they leave artists without clear protection. Any legal challenge would depend on courts extending existing doctrine to a context the law never anticipated. 

After the prompt: 

Legal uncertainty does not end once developers have trained an AI model; uncertainty shifts to the outputs and to the users who shape them. As creators put time and effort into directing AI tools, the law must decide whether the result qualifies for copyright protection. The key question is whether using AI amounts to authorship… or to something less. 

Théâtre D’opéra Spatial illustrates how unsettled the law remains. Jason Allen generated over 600 text prompts to guide Midjourney’s output. Then, he edited the result in Photoshop and upscaled it using Gigapixel AI. The final image reflects both aesthetic judgment and technical refinement. Still, it remains unclear whether that process crosses the legal threshold for authorship. 

Under section 5(1) of the Copyright Act, copyright protects only “original” works. In CCH v. Law Society of Upper Canada, the Supreme Court held that originality requires skill and judgment, meaning intellectual effort rather than something that can be “characterized as a purely mechanical exercise”. Although the Act does not explicitly require a human author, Canadian law assumes it. As a result, courts are unlikely to protect work generated entirely by AI. 

The uncertainty lies in the middle. When a person selects prompts, refines outputs, and adds further edits, that process may satisfy the CCH standard. But the line between authorship and assistance is thin, and without legislative reform, courts must draw it on a case-by-case basis. 

Legal consequences:  

Copyright law struggles to adapt to the use of AI in art, yet it remains the only standard by which courts and agencies may judge the legal entitlement of creators. When artists disclose how they make their work, they risk losing protection if they have employed AI. In a system that recognizes only human authorship, silence has become a strategy. 

In Thaler v. Comptroller-General of Patents, a U.K. court rejected a patent application for an invention that DABUS, an AI system, produced. Stephen Thaler refused to name a human inventor because none existed. The court noted that Parliament had not anticipated non-human inventors, but still enforced the statute as written. Thaler lost protection. 

Jason Allen made a different choice. In 2022, he applied for U.S. copyright in Théâtre D’opéra Spatial without revealing that Midjourney had generated the image. The Copyright Office, already aware of the controversy, investigated. Allen declined to disclose the extent of his post-prompt touch-ups and was denied registration.  

These cases expose a system shaped by outdated assumptions. Until lawmakers revise how copyright and patent systems define creative work, creators will continue to choose between honesty and protection. 

Solutions: 

The law does not need to treat AI systems as authors, but it must recognize how humans use them to create. Creators are already putting time, skill, and judgment into shaping AI-generated work. The current framework offers them no clear protection. Going forward, lawmakers can limit copyright to human authors, expand the definition of authorship, or create a new set of rights for AI-generated content. 

The first option draws a firm line. Only humans can claim copyright. If no human meets the originality threshold, the work enters the public domain. This keeps the law consistent, but it leaves creators exposed, even when their involvement is substantial. 

The second option redefines who counts as an author. It assigns rights to the person who directed or arranged the work, even if they did not make any creative decisions. That might include prompt engineers, dataset curators, or editors. This approach expands coverage, but risks awarding authorship where no real authorship occurred. 

The third path avoids that problem. Lawmakers could create a separate set of rights for AI-generated works. These rights would not depend on identifying a human author. Instead, they would grant limited economic protection to the person who shaped the output. Copyright already works this way in other domains, such as sound recordings. This model protects creative investment without requiring the law to treat AI-generated work as human authorship. 

Of these, the third offers the most realistic and flexible solution. It avoids stretching the meaning of authorship, but still protects the work that creators put into AI-assisted work. Just look at the piece itself:  Théâtre D’opéra Spatial sure does look like “real art.” The law does not need to call Jason Allen the author, but it should recognize that what he made deserves protection.  

The opinion is the author's, and does not necessarily reflect CIPPIC's policy position.