Mili Law Firm

Who Owns AI-Generated Content? China's 2026 Legal Framework Explained

July 8, 2026 · Copyright · AI Law · APNaji Patent / Mili Law
You generated an image with Midjourney and posted it on your company blog. Is it yours? You wrote an article with ChatGPT — who owns the copyright? These aren't hypothetical questions. In 2026, courts are delivering real answers.

Why This Matters Now

The global AIGC market surpassed $80 billion in 2025. In China alone, over 30 million pieces of content are AI-assisted or AI-generated daily. Yet the legal framework for AI-generated content copyright remains under construction worldwide. What we do have are landmark rulings — and they're already shaping how businesses must operate.

The "Spring Breeze Case": China's First AI Copyright Ruling

In late 2023, the Beijing Internet Court delivered China's first ruling on AI-generated image copyright. The court held that copyright protection depends on "human intellectual input." The plaintiff had used Stable Diffusion with detailed prompts, parameter adjustments, and iterative selection — the court found this process embodied sufficient personal expression and creative choice. The AI was merely a tool, and the resulting image was a copyright-protected work.

Key holdings:

Developments in 2025-2026

1. National Copyright Administration Guidance

In 2025, China's National Copyright Administration released draft guidance proposing a "three-factor test" for AIGC copyright protection: (1) originality of prompts — whether the instructions embody personalized creative intent; (2) human curation — whether selection among multiple outputs involved aesthetic judgment; (3) post-generation editing — whether the output underwent substantive human modification. Meeting at least two factors supports a copyright claim.

2. Platform Terms: Read Them Before You Generate

PlatformCopyrightKey Risk
ChatGPT / OpenAIOutput belongs to userOutput may be similar across users
Midjourney (paid)Images belong to userFree tier may not grant copyright
Stable DiffusionDepends on model usedOpen-source models — training data copyright unclear

3. Training Data Risk

When you upload client contracts, technical documents, or business plans to an AI tool for "smart analysis" — where does that data go? Some platforms' privacy terms allow using user inputs for model training. This is a rapidly growing area of inquiry for copyright litigation attorneys in China and globally.

Five Compliance Actions for Your Business

  1. Define AI use boundaries: Create an internal AI policy distinguishing "AI-assisted" (human-driven) from "AI-generated" (machine-driven) content.
  2. Document the creative process: Save prompts, iteration history, and evidence of human editing — these are your strongest proof of authorship.
  3. Register key commercial outputs: Even AI-assisted marketing copy and designs can be copyright registered. Registration certificates carry strong evidentiary weight in Chinese courts.
  4. Audit platform terms: Specifically the data usage clause — is your input being used to train models?
  5. Keep core IP off public AI tools: Trademark designs, core technical documentation, and undisclosed business strategies should never touch public AI models.
⚠️ AI-assisted content. For attorney/agent review only. Platform terms are subject to change; check current versions.

Using AI to generate content? Unsure who owns it?
We'll audit your risks and handle copyright registration.

📋 Free AI Compliance Checklist