AI and Intellectual Property: Who Owns AI-Generated Code or Content?

Introduction

Generative artificial intelligence has propelled the professional world into an era of rapid innovation. By 2024, nearly 70% of companies have already tested or implemented generative AI tools, whether for producing code, writing marketing content, designing visuals, or automating workflows. This rapid democratization raises fundamental questions, often overlooked in the initial enthusiasm: who is the legal owner of AI-generated code or content? This crucial question for protecting your digital assets remains a blind spot for many entrepreneurs and executives. Ignoring it can lead to complex disputes, significant financial losses, and damage to your brand's reputation.

As a technical partner specializing in custom application development and AI automation, Aetherio helps its clients anticipate these risks. Based on our experience with critical projects, we understand that technical mastery alone is no longer enough: a strategic vision integrating legal aspects is essential. This article aims to decipher the subtleties of AI and intellectual property, in light of French and European regulations, the Terms of Service (TOS) of leading AI tools, and best practices for securing your creations. For more details on intellectual property in web development in general, you can consult our guide on intellectual property in web development.

The Legal Status of AI-Generated Content and Code in France and the EU

In Europe and France, copyright, as codified in the Intellectual Property Code, is intrinsically linked to the notion of original human creation. For a work (literary, artistic, software) to be protected by copyright, it must bear the imprint of its author's personality, reflecting their intellectual and creative choices. This is precisely where generative AI introduces a gray area.

Human Contribution: A Prerequisite

The general principle is clear: content or code entirely generated by artificial intelligence, without significant and original human intervention, cannot, under current French and European law, be protected by copyright. AI is considered a tool, not a creator in the legal sense of the term. Consistent case law requires a work to emanate from the intellect of a physical person.

This means that a simple prompt like "Generate a Python code for a sorting function" or "Write a paragraph about the benefits of AI," where the AI does all the work, renders the result unprotected by copyright. Anyone could reuse this content without risking infringement prosecution.

The Nuances: When AI Becomes a Creative Aid

However, the reality of generative AI use is rarely so simple. In most cases, AI acts as a creative co-pilot. The user intervenes at several levels:

1. Prompt Design: An elaborate, precise prompt that guides the AI toward a specific artistic or technical direction, incorporating stylistic or functional constraints, can be considered an expression of human creative choice. This is the essence of "prompt engineering," which is becoming a skill in its own right.
2. Selection and Curation: AI can generate multiple versions. Choosing the best one, adapting it to a specific context, falls under an artistic or functional human decision.
3. Modification and Improvement: Very often, the raw output of AI is reworked, parts are rewritten, others are removed, elements are added manually. It is this "human added value" that, depending on the importance and originality of the contribution, can grant the final result copyright protection.

In summary: Just because AI is used does not mean copyright disappears; its scope focuses on the ingenuity and human personalization of the final result. The proportion of human input is critical. Content where AI serves only as an iteration/suggestion phase before substantial human rewriting has a much better chance of being protected than "click & generate" content.

Terms of Service (TOS) for AI Tools: A Mandatory Read

Beyond the general legal framework of intellectual property, a crucial dimension is often overlooked: the Terms of Service (TOS) of the generative AI platforms and tools you use. These documents, despite their often tedious nature, are the actual rules of the game determining who holds the rights to the generated outputs and, equally critically, who can reuse your inputs (your prompts, your working data).

Key Provisions of Major AI Players' TOS

Each AI provider has its own TOS, and these can evolve rapidly. It is imperative to stay informed of the latest versions, as they can have major implications for your company's intellectual property.

• OpenAI (ChatGPT, DALL-E) & Anthropic (Claude): Generally, these platforms often grant the user a non-exclusive, worldwide right to the generated content. This means the user can use, modify, and sell the content. However, the most sensitive clause often concerns the reuse of your prompts and outputs for improving their models. These companies may reserve the right to use your data (anonymized or not) to train their AIs, raising questions of privacy and ownership of training data. Make sure to use their "Enterprise" versions or dedicated APIs, which often offer stricter guarantees against data reuse.
• Cursor, Lovable, GitHub Copilot (for code): These tools have become indispensable for AI-assisted development. They pose the thorny question of ownership of the generated code. While TOS may grant users rights over the generated code, the real danger lies in the code's origin. Was the model trained on proprietary or open-source licensed code? We will address this point in more detail later. The article on the evolution of AI-assisted development provides a good introduction to the world of Vibe Coding.

The Risk of Dual Ownership

A common scenario occurs when the AI tool retains certain usage rights over your output (to improve its model), while you, as the user, believe you are the sole owner. This potential "dual ownership" can be problematic if the generated content is strategic for your company or contains confidential information.

Consequences of Ignoring TOS

Not reading and understanding the TOS can have serious consequences:

• Loss of Trade Secrets: If your prompts contain confidential or strategic information and the AI provider reserves the right to reuse them to train its model, your future competitors could indirectly benefit from your innovations.
• Ownership Disputes: If you market AI-generated content or software whose usage rights are debatable due to TOS or lack of significant human input, you expose yourself to lawsuits.
• Violation of Others' Property: AI can generate code or content that, unbeknownst to you, is too close to existing copyright-protected material. TOS do not protect you from this risk; they can even make you liable.

Aetherio Recommendation: Before adopting a new generative AI tool in your company, designate a person or team to carefully read and summarize the TOS, identifying critical clauses regarding intellectual property, data confidentiality, and input/output reuse. This should be integrated into the tool validation procedures, just like security or GDPR compliance (for which we have an article on GDPR compliance for data used by AI).

Concrete Risks: Involuntary Infringement and Hidden Licenses

Beyond the question of who "owns" AI-generated work, there is an even more insidious problem: the risk that this work, even with substantial human input, too closely resembles an existing work protected by copyright or specific licenses.

AI-Generated Code and the Challenge of Open Source Licenses

Generative AI code models (like GitHub Copilot, Cursor, Lovable) are trained on massive amounts of existing code, including open source code. The problem: this open source code often comes with licenses (MIT, GPL, Apache, etc.) that impose conditions for use, modification, and distribution. Our guide on open source licenses covers these aspects in detail. Here are the main risks:

• Inclusion of Code under Restrictive Licenses: AI could generate a code snippet that closely resembles code under a GPL (General Public License), which stipulates that any derivative code must also be open source and GPL licensed. If your company incorporates this code into its proprietary software unknowingly, it risks having to open source its own code or face lawsuits for license violation.
• Lack of Attribution: Many open source licenses require clear attribution of the original author. AI, by nature, does not provide these attributions, leaving you with no way of knowing where each line of code it suggests "comes from."
• Introduced Security Vulnerabilities: Generated code can be functional, but it is not free of vulnerabilities, hence the crucial importance of systematic human review. Our article on the security flaws of vibe coding explores these risks.

Concrete Cases and Jurisprudence

While case law is still developing, several lawsuits have already been filed. GitHub Copilot, for example, has faced complaints regarding the use of open source licensed code without attribution. These cases highlight the need for increased vigilance.

AI-Generated Textual/Visual Content and Infringement

The same principle applies to textual, visual, or audio content:

• Involuntary Plagiarism: AI trained on a corpus of existing texts could generate a paragraph, a sentence, or even an idea formulated so closely to a protected work that it would constitute infringement. This is particularly risky for marketing texts or blog articles if, for example, you use an automated AI content creation service.
• Visual Resemblance: For generated images, the risk is a disturbing resemblance to existing artistic works or logos. This can lead to accusations of copyright infringement or trademark violation.

Legal Liability: The User on the Front Line

It is crucial to understand that, generally, liability will fall on the company or individual who uses and publishes the AI-generated content or code. The AI provider (OpenAI, Google, etc.) will often disclaim this responsibility via its TOS, arguing that it is a tool and the user is solely responsible for validating and publishing the content. This is why a proactive approach is essential to avoid disputes. Managing risks related to unsupervised use of generative AI, or Shadow AI, is now a major challenge for businesses.

Best Practices for Serene Use of Generative AI

Given this complex legal landscape, inaction is not an option. Adopting rigorous best practices is essential to minimize risks and maximize the benefits of generative AI.

1. Document Human Input and Processes

This is the first line of defense to prove originality and human creative input:

• Retain Detailed Prompts: Record the precise prompts and the "prompt engineering" you implemented to guide the AI. This demonstrates your creative intent.
• Archive Different Iterations: Keep track of the various versions generated by AI and the manual modifications you made. Highlight the added value of human review and refinement.
• Track Time Spent and Decisions Made: Document the time a physical person spent refining, restructuring, or rewriting the content. This strengthens the argument for human contribution.

2. Choose AI Tools and Their TOS Wisely

Don't rush to use the first tool you find. Evaluate them based on their usage licenses:

• Read TOS Carefully: Implement legal monitoring of the TOS for the tools you use. Favor "Enterprise" versions or APIs that often guarantee better protection of your data and outputs.
• Opt for Enhanced Guarantees: Some providers offer more favorable intellectual property clauses, or even indemnities in case of disputes due to plagiarism generated by their AI. This is becoming increasingly common.
• Diversify Sources: Don't rely on a single tool. Using different AIs for the same task can help diversify content origin and reduce the risk of repeating specific models.

3. Systematic Human Review of AI Outputs

AI is an assistant, not a substitute:

• Critical Review and Rewriting: Every AI-generated element—whether code, text, or visual—must undergo critical review and rewriting by a human expert. This is your chance to add your "touch," your originality, and verify factual accuracy (AI can "hallucinate").
• Plagiarism and License Verification: Use anti-plagiarism tools for texts and, for code, dependency and license analysis tools to identify any code snippets under restrictive licenses. This is essential for ensuring compliance with open source licenses.
• Legal Consultation: For sensitive, strategic, or high-stakes commercial projects, consultation with an intellectual property attorney is essential. We are not lawyers, but we can direct you to experts.

4. Team Awareness and Training

Your employees are the front line of AI adoption. They must be trained:

• Clear Guidelines: Establish internal guidelines on the use of AI tools, types of data never to submit (confidential information, sensitive personal data), and output validation procedures.
• Accountability: Every user must understand that they are responsible for the content they generate with AI and publish under their own name or the company's. A good software development contract usually includes clear clauses on intellectual property rights.

Conclusion

Generative artificial intelligence is a transformative force for businesses, capable of accelerating innovation, optimizing costs, and increasing productivity. However, enthusiasm for these technologies should never overshadow the need for rigorous legal vigilance, particularly concerning the intellectual property of generated code and content.

To recap the key points:

1. Human input is vital: In France and Europe, copyright prioritizes originality and the imprint of human personality. A 100% AI output is not protectable.
2. TOS are contracts: Carefully read the Terms of Service for each AI tool you use, as they define what can be done with your prompts and outputs, and may imply reuse by the provider.
3. The risk of infringement is real: AI can generate content or code that too closely resembles works protected by copyright or subject to specific licenses (especially open source), incurring your liability.
4. Prevention through documentation and processes: Documenting your human input, choosing tools with transparent and favorable TOS, and implementing systematic human review are essential actions.

At Aetherio, we believe that sustainable innovation requires informed risk management. As an outsourced CTO and strategic partner, we don't just build your future business application or MVP; we also guide you through complex topics like this, integrating a technical, business, and legal vision from the design stage. Your success relies on the strength of your foundations, including legal ones. Feel free to contact us to discuss your projects and secure your innovation.

Further reading:

• Intellectual Property in Web Development
• Shadow AI: The Invisible Risk of Your Teams Using ChatGPT Unsupervised
• Vibe Coding: Cursor, Claude Code, Windsurf... How AI is Revolutionizing Development in 2026