Copyright

Copyright Infringement Liability for AI-generated Content

The Intellectual Property Department (IPD) has outlined in its Copyright and Artificial Intelligence Consultation Paper that when an AI-generated work infringes copyright, liability rests with the individual or entity that made the necessary arrangements that caused the AI to infringe. This principle is consistent with how liability is determined for non-AI-generated works.

Example 1
If an AI-generated work infringes copyright without any prompts from the user suggesting such copying, and the AI developer has the ability to prevent this infringement, the developer should be primarily liable. However, the user may also be held accountable if they copy or publicly share the infringing AI-generated work.

Example 2

Conversely, if a user’s prompts explicitly indicate the desire for the AI to produce an infringing copy of a copyrighted work, the end-user could be held liable for copyright infringement alongside any potential liability of the AI developer.

The IPD plans to issue interpretive guidelines that will clarify how existing legal principles for copyright infringement apply to cases involving AI-generated works. These guidelines will include practical suggestions and specific examples.

Contractual arrangement between AI system owners and end-users

You should also carefully review the terms of use or service agreements, particularly those addressing liability related to copyright infringement from AI-generated works.

Some AI system owners may include clauses that limit their liability to users. In these cases, users might be required to indemnify the AI system owners against any third-party claims resulting from the use of AI-generated works. Conversely, some AI system owners provide protections for users, covering legal costs and damages from copyright claims to foster trust in their AI services.

Ingesting Copyrighted Works as Training Data 

The integration of copyrighted works into AI training data raises important legal and ethical concerns, particularly when the outputs compete with original works. Critics argue that this practice undermines the legitimate interests of copyright owners, who deserve compensation for the use of their creations. On the other hand, AI developers assert that Gen-AI enables innovative reuses of data embedded in copyrighted works, claiming that these transformative applications can qualify as fair use.

To address the balance between copyright holders and users, many countries have introduced a Text and Data Mining (TDM) Exception. Below is an update on the proposed exception in Hong Kong.

The Coming Text and Data Mining (TDM) Exception in Hong Kong

To support the development of AI models that require extensive use of copyrighted materials, the Hong Kong government plans to amend the Copyright Ordinance.

The proposed TDM Exception will allow the copying of copyrighted works for computational data analysis and processing, specifically for the development, training, and enhancement of AI models, without needing licenses from copyright owners. This exception will apply to both non-commercial and commercial uses, fostering the growth of AI technology.

TDM is not limited to AI development; researchers also use TDM to analyze large volumes of digital resources, aiming to acquire new knowledge, advance research, and uncover novel patterns and trends.

The proposed exception will include specific restrictive conditions to balance the interests of copyright owners and users. Key conditions include:

The Intellectual Property Department (IPD) will formulate codes of practice or guidelines to assist with the implementation of feasible opt-out options, as well as provide guidance on record-keeping and disclosure.

Ingesting Open Access Works under CC Licenses

Note that using open-access works licensed under Creative Commons (CC) licenses as training data is not always permissible. Some licenses impose restrictions such as non-commercial (NC), no-derivatives (ND), or share-alike (SA), which require more than just attribution.  

The following table summarizes the usage of CC-licensed works as training data, assuming that the AI-generated output will be shared publicly and that copyright permission is required (e.g., no TDM exception):

For more information, please refer to Using CC-Licensed Works for AI Training and the official flow chart.

Friendly Reminder: Do Not Upload HKU Library E-Resources to Third-Party Platforms 

Most e-resources provided by the HKU Libraries are governed by license agreements with publishers and vendors. These agreements often prohibit the uploading of content to third-party platforms, including Gen-AI tools, even for educational purposes. Non-compliance with these license terms may constitute copyright infringement, resulting in liability for any damages incurred. 

Licensing Scholarly Content for AI Training

Many academic publishers are either announcing or negotiating licenses to use their scholarly content as training data for large language models (LLMs). They claim that this licensing can enhance the accuracy and relevance of AI models, reflecting a commitment to ensure authors' ideas make the fullest possible contribution.

To keep track of these developments, Ithaka S+R has launched the Generative AI Licensing Agreement Tracker, which documents public agreements and analyzes their impact and underlying strategies. It is important to note that this tracker only includes publicly disclosed agreements, and there may be additional undisclosed deals.

In response to concerns about the use of scholarly content without researchers' knowledge, some publishers are adopting an opt-in approach, actively seeking permission before licensing their content for AI training. While the decision to participate is ultimately personal, consider the following reasons for opting in or opting out.

Understanding Your Rights in Publisher Agreements for AI Training and Usage

When publishing your work, it’s crucial to understand your rights concerning AI traing and usage. Start by carefully examining your contracts to determine what you have authorized your publisher to do with your work. Pay close attention to the following areas:

1. Licensing and Royalties: Identify any terms related to royalties and shares of licensing revenue. Make sure you understand how your work can be used and how you will be compensated.
2. Subsidiary Rights: Look for clauses regarding subsidiary rights, which may permit publishers to license your work for additional purposes. You can negotiate to retain these rights or establish a profit-sharing arrangement for any licensing deals.
3. Publisher Practices: Stay informed about your publisher’s practices regarding AI licensing. If you find the information in your contracts unclear, don’t hesitate to ask questions or seek clarification.
4. Rights Assignment: Be aware that if you have assigned all rights to your publisher, they have the authority to license your work for various purposes, including AI training, and can profit from those deals without your consent.
5. Negotiation for Future Works: For your upcoming publication, consider attaching an author addendum to your agreement that outlines your preferences for AI usage. The Authors Guild offers a model clause that prohibits using your work for AI training without your explicit permission.

CC Signals: A New Initiative from Creative Commons

CC Signals is a groundbreaking framework that empowers content owners to articulate their preferences for how their content can be used in AI training. This initiative seeks to strike a balance between the needs of creators and the demands of AI development, fostering reciprocity, shared benefits, and openness.

CC signals are designed to be interpretable by both machines and humans. The four signals proposed by Creative Commons are:

Signal		Description	Proposed Combinations
	Credit	Ensure appropriate credit is given based on the method, means, and context of use.	Credit
	Direct Contribution	Provide monetary or in-kind support to the content owner for the development and maintenance of assets.	Credit + Direct Contribution
	Ecosystem Contribution	Provide monetary or in-kind support back to the ecosystem benefiting from the use of the content.	Credit + Ecosystem Contribution
	Open	The AI system must be open, adhering to standards like the Model Openness Framework or Open Source AI Definition.	Credit + Open

Credit is a fundamental component of each combination, emphasizing reciprocity and benefiting the broader knowledge cycle. For more information on the signals, click here.

Legal and Ethical Considerations: While CC Signals are not a replacement for copyright licenses, they introduce an ethical layer to content sharing. Participation from AI developers is voluntary but is encouraged through community norms and reputational accountability.

Current Status: CC Signals is in the public feedback phase, with plans to launch an alpha version in November 2025. This initiative primarily targets large dataset holders rather than individual creators.