Generative AI, represented by OpenAI’s ChatGPT, is reshaping the way we produce content worldwide. From academic essays and commercial contracts to news reports, computer code, and artwork, AI-driven writing and creation tools have grown exponentially within just a few years. They have significantly boosted productivity, lowered the threshold for creation, and opened new economic opportunities for businesses and individuals.
However, behind this rapid development, intellectual property (IP) disputes have become a global concern. Were copyrighted works used lawfully in AI training? Can AI-generated content be protected by copyright? Do users bear liability when they unknowingly publish infringing outputs? These questions are not only legal puzzles but also critical issues for the future of innovation ecosystems and economic incentives.
This article explores these challenges from five perspectives:
How ChatGPT learns and why this raises legal disputes
Legal risks for AI users
The copyright ownership dilemma of AI-generated works
Global legal and regulatory trends
Potential solutions and economic implications for the future
ChatGPT’s language capabilities come from training on massive datasets. These datasets include books, news articles, websites, and social media posts. The problem: much of this material may be copyrighted, and authors often did not grant explicit permission.
Unauthorized Use: Using protected works without a license could amount to copyright infringement.
Fair Use Ambiguity: U.S. law allows “fair use” for purposes such as teaching and research. But whether AI training qualifies as “transformative fair use” remains unsettled.
The Black Box Problem: Since AI models are statistical black boxes, it is almost impossible to trace a specific output back to an identifiable training text—making enforcement difficult.
Authors Guild v. OpenAI (USA, 2023)
The Authors Guild filed suit against OpenAI, claiming their works were used without permission for training ChatGPT. This case could set an important precedent for copyright boundaries in AI training.
Getty Images v. Stability AI (Europe, 2023)
Getty Images accused Stability AI of illegally using millions of copyrighted images to train its models. The case was settled, highlighting the importance of licensing in AI commercialization.
Even users with no intention to infringe may face legal risks when publishing AI-generated content:
Unintentional Copyright Copying: AI outputs may closely resemble existing works, creating “accidental plagiarism.”
Unclear Liability Distribution: Laws are still vague about whether the AI platform or the user bears primary responsibility.
Terms of Service Risks: Many AI companies shift liability to users via their terms of service, making users the “first line of defense” in lawsuits.
Lee v. AI Generated Articles (California, 2024)
A user submitted AI-generated news articles that strongly resembled existing reports. The court ruled that users have a duty to conduct preliminary review of AI outputs.
This signals an emerging legal trend: even if the platform provides the tool, users may still be directly liable.
Most copyright systems worldwide require human authorship. This leads to three key outcomes:
Fully AI-Generated Works: Typically not eligible for copyright.
Hybrid Models: When a human contributes significantly through prompts or edits, copyright may be recognized.
Generic Prompts Problem: Simple prompts like “write a Shakespearean sonnet” are unlikely to establish originality.
If AI-generated works cannot be copyrighted, they may be freely copied and reused. This has two consequences:
It undermines exclusive commercial monopolies of creators.
But it broadens public access to knowledge, benefiting education, research, and social welfare.
U.S. Copyright Office – Thaler Decision (2022)
Denied copyright for a fully AI-generated artwork, affirming the human authorship requirement.
EU AI Act (Draft 2024)
Requires transparency in AI training and outputs but does not grant copyright protection to AI-generated works.
Countries are adopting diverging approaches to AI governance.
Multiple lawsuits against OpenAI and Stability AI are ongoing.
Court decisions will determine whether large-scale AI training can be shielded under “fair use.”
The EU AI Act classifies AI systems into risk levels, requiring transparency and compliance audits for high-risk systems.
It emphasizes user rights and platform obligations.
Japan: Pursues flexible fair use policies to encourage AI innovation.
China: Revised its copyright law in 2024, mandating AI platforms to ensure data legality and strengthening compliance for AI-generated commercial content.
| Case/Regulation | Region | Issue | Status |
|---|---|---|---|
| Authors Guild v. OpenAI | USA | Unauthorized training | Ongoing |
| Getty Images v. Stability AI | Europe | Image copyright | Settled |
| Lee v. AI Generated Articles | USA | User liability | Partial rulings |
| U.S. Copyright Office – Thaler | USA | Copyright for AI works | Denied |
| EU AI Act (Draft 2024) | Europe | Transparency requirements | Pending |
To strike a balance between innovation and protection, policymakers and businesses may adopt the following strategies:
Introduce licensing schemes similar to the royalty system in the music industry, where creators receive compensation when their works are used for training.
Differentiate between non-commercial research/educational use and large-scale commercial training, with clear judicial guidance.
Grant limited copyright or neighboring rights to AI-generated works—protecting creators’ interests while maintaining public accessibility.
Legislation should specify:
Platforms are primarily responsible for ensuring training data compliance.
Users are secondarily responsible for conducting reasonable review of outputs.
This dual system would both protect users and incentivize companies to adopt transparent data-use practices.
ChatGPT is not only a technological breakthrough but also a stress test for intellectual property law. In the short term, legal uncertainty may accelerate innovation. But in the long run, unresolved IP boundaries risk undermining creative incentives and slowing adoption of generative AI.
The future of AI regulation hinges on three pillars:
Clear rules for data usage in AI training.
A workable rights framework for AI-generated outputs.
A fair distribution of liability between platforms and users.
Only by balancing innovation and protection can generative AI truly drive economic growth and social progress—rather than becoming a legal minefield.
