Copyright is an important factor to consider when using AI tools or content to support your study, research, or teaching.
As courts and policymakers continue to address the legal and technical issues surrounding intellectual property (IP) and copyright in the development and use of GAI, it can be challenging to determine whether your use of these tools is both legal and ethical.
The following outlines some of the copyright concerns and gives recommendations for managing the risks and using GAI tools responsibly.
RISK: Prompt material is ingested and reused by the AI model, thereby breaching copyright or licensing terms.
Most AI platforms require users to confirm (warrant) that they have permission to upload content. If it is not your own work, check the copyright and licensing agreement before uploading it.
Some platforms require users to grant the AI developer a licence to use uploaded content for refining or training the model. Sublicensing third party material without permission may infringe copyright or breach a licence agreement.
Using a link to material that is legitimately freely available is generally permitted.
When assessing or using GAI tools, check for the 'Terms of Use' in relation to 'user content' or 'input data' and 'results' or 'output data'.
Note: Privacy policies may say "your data is not used for training". In a privacy policy, 'your data' usually refers to any personal information that a company collects about you. eg. your name, account info, usage data. A privacy policy may not include uploaded content or input data you provide.
Below is a summary for some frequently used platforms. In all cases users must have the right or permission to upload their content. Licensing content for reuse by the platform varies and in most cases users own or are responsible for outputs.
Terms of use/service may be revised without notice. This information was compiled in May 2025.
AI tool |
User licences inputs for reuse by the platform | Ownership of outputs | Terms of use and other information |
---|---|---|---|
Microsoft Copilot - |
Sublicensing not required. Not trained on user prompts. Stores files for 30 days. Does not use files for training.
|
Claims no rights in outputs. Microsoft undertakes to defend users against third party copyright claims in outputs if infringement was not intentional. |
|
Microsoft Copilot - |
User grants Microsoft the rights to content and the rights to sub-licence to suppliers. Copilot is not trained on uploaded files. |
Claims no rights in outputs. | |
OpenAI ChatGPT, |
User retains ownership of inputs. OpenAI may use content to provide, maintain, develop and improve services. Users can opt out of training. |
User owns rights (if any) in outputs, but not if it includes third party content. | |
Anthropic Claude 3.7 Sonnet |
User retains rights to inputs. Model is trained on inputs if user provides feedback of if input is flagged for policy violations. |
User owns rights (if any) to outputs. | |
Gemini
|
User retains ownership of IP in content, but licenses Google to use it to provide and develop their services. Gemini is integrated into all Google products. |
Google does not claim ownership of generated content. |
Google's Terms of Service are very broad and cover most of their diverse array of interrelated services. |
Notebook LM |
Google NotebookLM claims not to use your personal data, including your source uploads, queries, and the responses from the model for training. Use is under Google's Terms of Service. |
Google does not claim ownership of generated content. | |
Perplexity AI Perplexity |
User grants license to Perplexity to use content to provide and improve the service and publish and distribute output based on the input. | "You must not use the service to generate output in violation of IP. You may not publish any outputs without clearly citing the services". | Perplexity AI Terms of Service |
Human Intelligence Technologies Research Rabbit |
You automatically grant all rights to use your contributions for any purposes in any media and you waive any right of attribution. | You are licensed to download and print copies of the content for personal use. | Research Rabbit Terms and Conditions |
Napkin AI Napkin |
You retain ownership of content you create or upload. You grant Napkin AI a licence to use your content to provide and improve the app. The licence continues when you stop using the service. Data may be used for improving the model. |
Not addressed. Napkin claims they do not use or incorporate images images scraped from the internet, ensuring visuals you receive are free from copyright concerns related to existing internet content. |
Napkin Terms and Conditions |
Gamma Tech Gamma |
You retain ownership of your content. You grant Gamma Tech a licence to your content for providing services to you and other registered users. Content submitted to the public area is visible to others. You can manage access settings. |
Not addressed, however Gamma claims no responsibility for output results. | Gamma Tech, Inc. Terms of Use Agreement |
Eduaide Ai LLC Eduaide.ai |
You retain ownership of your content. You grant Eduaide.Ai a limited licence to use content posted by you in order to provide the services to you. |
Not addressed, however claims no responsibility for AI model outputs. | Eduaide.Ai Terms of Service |
Canva Magic Studio Products |
You retain ownership of your content. You grant Canva a limited licence to use your content for providing their service to you. You acknowledge that content may be shared with third party AI model providers in order to provide the Canva service. Your content may be used to improve AI unless you opt out in your account privacy settings. |
Canva claims no rights in outputs. The licence and acknowledgements you give for inputs also apply to your outputs. |
Canva Content License agreement
|
RISK: False attribution resulting from reliance on AI search engine chatbots.
AI is integrated into a number of search engines and their chatbots offer alternatives to traditional keyword searches. Instead of scrolling through multiple pages in search results, you can get a top level summary that seems to understand the nuances of natural language. The best ones will even include links to their sources.
Even if an AI summary includes links to real, rather than hallucinated references, the information it is quoting may not come from that reference or may be misinterpreted. Do not rely on AI tools for correct citation. Incorrectly attributing an author, repeating an out of context phrase, or stating that someone said something they did not say are all examples of infringing an author's moral rights.
It is reasonable to use AI search engines for finding information, but you should not quote AI outputs directly. Instead, trace back and check the original source/s before using it:
RISK: Potential lack of IP protection for creative and research outputs where GAI has been used.
The copyright status of GAI outputs varies by jurisdiction. In Australia, human authorship is a prerequisite for copyright protection. The United States follows a similar approach; works not controlled by a human cannot be registered for copyright. Even highly creative prompts typically do not provide the level of control over the GAI’s processing necessary to meet the threshold for copyright registration.
In contrast, the UK and New Zealand recognize copyright in computer-generated works. In these jurisdictions, the 'author' is defined as the person who makes the arrangements necessary for the creation of the work. However, ambiguity remains over who this person is: the individual prompting the GAI, the creator of the model, or those who trained it.
While most jurisdictions require human authorship for copyright, protection may still be available when a person contributes significantly to the creative process.
According to the Copyright Agency (August 2024), “Australia does not have a copyright registration system. If someone were sued for breach of copyright for using an AI-generated output created by someone else, a court would likely consider the level of human involvement in generating the output and how it was used by the alleged infringer.”
The Report on Copyright and Artificial Intelligence Part 2 issued by the US Copyright Office (29 Jan 2025) indicates there is potential for copyright protection in the following circumstances:
To manage copyright and ownership risks when using GAI:
For outputs that are predominantly AI-generated, consider applying a Creative Commons Zero (CC0) public domain dedication provided you are sure they do not contain infringing content (Creative Commons,18 Aug 2023). Regardless of the level of human input, always acknowledge the use of AI tools.
RISK: Plagiarism and/or infringing third party copyright in GAI outputs.
GAI models are trained on vast amounts of content (data) available over the internet. Some of the content is public domain (free of copyright), but most is protected by copyright. In many cases the content has been used without permission or licences, with no attribution or financial compensation to the copyright creators and owners. This has resulted in high profile copyright lawsuits against AI developers and a great deal of anxiety for individuals whose moral rights (right to attribution) and economic rights (right to control how their work is used) have been infringed.
Copyright laws seek to balance the rights of creators and owners with the right of public access to information, knowledge, and creative works. Research and creative outputs and innovations are built on and inspired by earlier work. Copyright protection encourages and supports creativity but overly strict copyright laws can limit opportunities for innovation and education. Courts and policy makers around the world have always grappled with the challenge of finding a fair balance as new technologies have developed, however lack of transparency in the training data and functioning of AI models adds to the current complications.
Proposed solutions to the issue with copyright infringement and AI training under consideration include:
Until solutions are implemented, users of AI tools need to assess the risks and make their own ethical judgements on how they use and share GAI outputs. If an AI tool generates material that is a copy or substantially similar to another work, it may be infringing the copyright on the original. Liability is unclear, but if you publish it, there is a risk of being held liable for infringement. Even if copyright is not infringed because it is not a clear reproduction of copyrighted content, there is a risk of plagiarism.
Recommendations:
RISK: GAI outputs may impact the creativity industry market.
GAI is capable of producing original material that does not infringe copyright or plagiarise. However, there is potential for GAI material to impact the market for human creativity. Artists, authors and musicians earn royalties through platforms such as Spotify, image stock libraries and YouTube. By using material generated with AI as background music or for illustration, people relying on royalty income may lose some of the market share. This 9News Article on 12 April 2025, 'This is taking away from me making a living' by Evie Hilliar illustrates the issue.
Recommendation:
Charles Sturt University acknowledges the traditional custodians of the lands on which its campuses are located, paying respect to Elders, both past and present, and extend that respect to all First Nations Peoples.
Charles Sturt University is an Australian University, TEQSA Provider Identification: PRV12018. CRICOS Provider: 00005F.