The European Union just drew a sharp line in the sand. If a machine made it, you cannot own it. The European Court of Justice and national copyright offices are aligning on a simple rule: copyright requires human handiwork. This is not a minor policy tweak. It changes how software teams, game studios, and digital agencies build products.
For years, companies hoped that complex prompts or expensive GPU runs would grant them exclusive rights to their assets. The EU has put an end to that hope. Purely machine-generated code and images belong to the public domain the moment they leave the model.
The Legal Foundation of Human Authorship
EU copyright law does not have a single code for AI. Instead, it relies on a long history of court decisions about what makes something original. The foundation rests on the concept of an author's own intellectual creation.
The InfoSoc Directive (Directive 2001/29/EC) harmonized certain aspects of copyright law across the EU. It did not define what a work is, leaving that to the courts. The Court of Justice of the European Union (CJEU) stepped in to fill this gap.
In the landmark Infopaq case (C-5/08), the CJEU ruled that copyright only applies to elements that express the author's personality. Later, in the Painer case (C-145/10), the court clarified that this means the creator must make free and creative choices to stamp the work with their personal touch.
When you run a prompt through a large language model or an image generator, you do not make those choices. The model does. The model selects the next token or diffuses the pixels based on statistical weights. The user merely sets parameters.
This distinction is critical. If you write a prompt asking for a database script, you are setting the parameters. You are not writing the code. The machine decides the loop structures and the variable names. Because the machine cannot hold legal rights, and you did not write the code, the output lacks copyright protection.
Why Prompts Do Not Count as Creative Choices
Many prompt engineers believe their work is similar to writing code. They argue that a prompt is a structured set of instructions that guides the model to a specific output. But this comparison fails.
A prompt is instructions. It is like telling a human painter to draw a blue horse. The person who gave the instruction does not own the resulting painting; the painter does. In the case of AI, the painter is a machine that cannot hold legal rights. Since the machine cannot own copyright, and the prompter did not execute the creative choices, the work falls into the public domain.
The German Federal Patent Court and regional courts in Europe have already applied this logic. They look at the causal link between the human mind and the final output. If the machine does the heavy lifting of expression, the link breaks.
Think of it as a camera. A photographer makes creative choices by selecting the angle and the lighting. The camera simply records the light. With AI, the user does not control the execution. The model uses random seeds and probability distributions to create the final output. The user cannot predict the exact placement of pixels or tokens. That lack of control is why courts reject the camera analogy for generative AI.
The Red Line of Human Intervention
This does not mean any work involving AI is automatically uncopyrightable. The law looks at the workflow. We can divide this into distinct phases.
First, the conception phase. This is where you plan the work.
Second, the execution phase. This is where the work is actually built.
If AI dominates the execution phase without human correction, copyright is lost. If a developer uses Copilot to generate a standard utility function, that function lacks protection. Relying heavily on these tools also introduces issues like context drift and review fatigue. But if the developer takes that code, rewrites parts of it, and integrates it into a larger system architecture, the overall system remains protected. The protection covers the human assembly and modifications. It does not cover the raw AI output.
The key is the amount of human modification. You must prove that your human choices shaped the final product. If you generate an image and spend hours manually painting over it in Photoshop, the final image is protected. The protection only covers your edits, not the base AI layer.
The Impact on Software Development
Software companies face a massive risk. Code bases are increasingly filled with AI-generated snippets. If a competitor can copy your code because it lacks copyright protection, your business model might suffer.
Consider database schemas or API integrations. If you generate a schema using a model, that schema might not have legal protection. A competitor could copy it directly without violating copyright.
This changes how we think about open source and proprietary licenses. To claim copyright infringement, you must first prove you own the copyright. If your codebase is mostly raw AI output, proving ownership becomes a legal nightmare. You have to isolate the human-written parts from the machine-generated parts.
And this is not just about code. It applies to documentation and system architecture designs. If these assets are generated without human intervention, they are free for anyone to copy.
Digital Assets and Game Design
Game studios use generative tools to create textures and 3D models. Under the EU ruling, these assets are vulnerable.
If a studio generates thousands of background textures using a diffusion model, those textures are public domain. A competitor can rip them directly from the game files and use them in their own commercial project. The original studio has no legal recourse under copyright law. This lack of protection is compounded by the rise of AI watermark removal tools, which make it easier to strip ownership markers from digital assets.
To prevent this, studios must implement hybrid pipelines. Human artists must take the AI-generated base assets and modify them. They need to paint over them or adjust geometry in ways that show clear human choice. The final asset becomes protectable because of the human modification, even if the base layer came from a machine.
This creates a new workflow. AI is used for rapid prototyping, but humans must do the final production work to secure the intellectual property.
Practical Strategies for Engineering Teams
Protecting your intellectual property in this environment requires clear rules and documentation.
First, track your tools. Keep a log of where AI is used in your pipeline. This helps you identify which parts of your product might be vulnerable.
Second, enforce human review and modification. Do not use raw AI outputs in production code or assets if those assets are critical to your competitive advantage.
Third, document the creative process. If you use AI to generate an initial draft, keep the version history showing how human engineers or designers edited and rewrote the final product. Git history serves as excellent evidence of human authorship.
Git History as a Legal Shield
Your version control system is your best defense. When you use AI tools, make sure your commits are small and frequent. This shows the progression of the code. It proves that a human developer reviewed, edited, and integrated the machine-generated snippets.
If you commit a massive file with 10,000 lines of code in a single step, it looks like raw AI generation. If you commit the code in small chunks, with clear commit messages explaining the changes, you build a record of human authorship. This record can be used in court to prove that the work is your own intellectual creation.
Fourth, rely on other legal protections. Since copyright is weak on AI outputs, use trade secrets and contract terms to protect your platform. You might not own the copyright to a generated dataset, but you can restrict access to it through a binding user agreement.
The Global Divergence
The EU is not alone, but its approach is particularly strict. The US Copyright Office has taken a similar stance, rejecting registrations for AI-generated images. However, the EU framework relies heavily on civil law traditions that focus on the personality of the author. This makes it even harder to argue for machine authorship in Europe than in common-law jurisdictions.
Companies operating globally must design for the strictest standard. If your workflow does not pass the EU test for human authorship, you should assume your assets are unprotected in Europe, regardless of what US or Asian courts decide.
This divergence means you cannot have a single global policy for digital assets. You must evaluate your IP protection country by country, focusing on where your primary market sits.
The Future of AI Contracts
We are seeing a shift in vendor agreements. When you buy software or hire agencies, you must ask about their AI usage.
Contracts now include clauses requiring developers to disclose if they used generative tools. If an agency delivers a logo or a code library that is purely AI-generated, they are delivering an asset you cannot own. You paid for exclusive rights, but you received public domain material.
Clients will demand warranties showing that all deliverables contain enough human authorship to qualify for copyright protection. Agencies must adapt by training their staff to use AI as a starting point, not a final destination.
This will change the pricing models for digital work. Purely automated generation will become cheap, while human-verified, copyright-protected work will command a premium.
Proving Authorship in Court
If you end up in a legal dispute, the burden of proof is on you. You must show that a human made the creative choices.
This is where your internal tool logs and version control systems become critical. If you can show a clear path from a blank screen to a finished product, with step-by-step human edits, you can defend your copyright. If you only have a prompt and a final output, you will lose.
Courts will look at the tools you used. They will ask if the tool allows for precise human control or if it relies on random generation. Tools that allow for manual adjustment, like digital brushes or code editors, help support your claim of authorship. Tools that generate complete files from a single click do not.
You should establish an internal audit process. Before publishing any critical code or design, verify that a human has made substantial modifications to the machine-generated parts.
The Shift to Technical Protections
Since legal protections are changing, companies must rely more on technical protections. This includes hardening infrastructure and securing AI platforms against unauthorized access.
If you cannot prevent competitors from copying your code using copyright law, you must make it harder to copy. This means using better obfuscation and server-side processing.
For data and content, you must use access controls. Paywalls and rate limiting become your primary lines of defense. If someone copies your data, you sue them for breach of contract rather than copyright infringement.
This shift will change how we architect applications. We will see more closed APIs and fewer open client-side applications as companies try to protect their uncopyrightable assets.
