Do You Own Copyright to AI‑Generated Creations? Legal Cases and Platform Policies Explained

AI is increasingly used as a creative tool in literature, art, and other fields, raising questions about who holds the copyright to the resulting works.

Case Studies

Domestic "AI Writing First Case" (2018)

In August 2018, Tencent Securities published a stock‑analysis article automatically written by the AI robot "Dreamwriter." Another website copied the article without permission, leading Tencent to sue for infringement and win the case, demonstrating that AI‑generated works can be subject to copyright protection.

Domestic First "AI Text‑to‑Image" Infringement Case (2023)

In February 2023, Mr. Li used Stable Diffusion to generate an image and posted it online. Mr. Liu removed the watermark and used the image in an article without permission. The court ruled in favor of Mr. Li, stating that the image qualifies as a "fine art work" with originality and is protected by copyright law.

The court’s reasoning included two points: (1) AI is a tool, not a legal entity, so it cannot own copyright; (2) the developers of Stable Diffusion do not claim rights over user‑generated content. Because Mr. Li contributed intellectual effort—conceptualizing the scene, crafting prompts, adjusting parameters, and selecting the final output—the work was deemed his, and his copyright was upheld.

Other Influencing Factors

Platform Agreements

Many AI services embed IP clauses in their user agreements. For example, Baidu's "Wenxin Yige" platform states that the intellectual property of content generated through the service belongs to the platform or its rights holders, effectively denying users ownership.

5.4 Users' content generated while using the service belongs to Wenxin Yige or related rights holders.

However, the platform also permits commercial use of the generated images within legal limits, which can be viewed as a licensing grant.

Midjourney’s terms claim that user‑generated works are the users' assets, but they also require a permanent license to Midjourney and list certain exceptions.

Training‑Data Material Disputes

AI models are trained on large datasets that may contain copyrighted material. If the training data itself is disputed, it could affect the ownership of the generated output, a topic the article notes for future discussion.

Key Takeaways

If an AI service’s user agreement explicitly assigns copyright of generated content to the service provider, the provider holds the rights.

If no such clause exists, the user who contributed creative input (e.g., prompts, parameter tuning) retains the copyright.

Readers are encouraged to read platform agreements carefully before using AI‑generated works in formal contexts.