How GenAI uses data
Copyright and protection of intellectual property
Privacy concerns

How GenAI uses data

As GenAI tools train, in part, based on the data that users input into them, it is important that you do not enter any personal data (your own, or others) or any copyrighted material because this information is not secure and can possibly become an output for another user. Beyond this, ethical concerns have been raised over the methods used for training these tools that include scraping internet data, as there is a lack of transparency around this in terms of use of copyrighted material, lack of consent, and the monetisation of the work of others without proper credit/payment.

For example, consider a situation where you are using a GenAI tool to help you write a blog. The GenAI tool is using databases of information taken from the internet. While its eventual output will seem like an original, unique piece of writing, it is based on previously inputted content, or content that formed part of the GenAI tool’s training. Who then owns the newly generated content? You, the user? The company that owns the GenAI software? The owner of the original content on which the GenAI tool was trained? Does copyright, patent, trademark infringement apply to AI creations? Is it clear who owns the content that GenAI platforms create for you?

This process is rife with legal risks, including intellectual property infringement. As GenAI is very much an emerging field, legal rulings and issues regarding policy are still in development.

Copyright and protection of intellectual property

The emergence of GenAI has sparked significant debate over intellectual property (IP) rights. A key concern is that AI models are often trained on copyrighted materials without explicit permission, raising legal and ethical questions about ownership and fair use. Artists, writers, and content creators argue that AI-generated works may infringe on their original creations, leading to unauthorised replication and potential loss of revenue. Additionally, determining authorship and copyright eligibility for AI-generated content remains a grey area, as current laws typically require human involvement for copyright protection. As AI continues to evolve, policymakers and legal experts are grappling with the challenge of balancing innovation with the protection of creators' rights, ensuring that AI-generated content does not undermine the value of human creativity.

Copyright infringement court cases

New York Times vs. OpenAI

The New York Times (and The New York Daily News and the Centre for Investigative Reporting) are suing OpenAI over the use of their subscription articles to train ChatGPT and CoPilot without their permission. They have illustrated cases where where these chatbots have output portions of these articles near-verbatim and argue that this is "substitutional" for the original work, competes with the original work and threatens high quality journalism. At the heart of this case is a debate over copyright "infringement" versus what can be considered "fair use" (regarding the unlicensed use of copyrighted materials for specific purposes, which OpenAI argues this falls under). You can read more about the case on NPR.

Several authors, including John Grisham and Jodi Picoult, have similarly taken out cases against OpenAI over use of their books for training data without permission.

Getty Images vs. Stability AI

Getty Images maintain curated databases of high quality images that are free to view but require a license for use. They are suing Stability AI over the use of images scraped from their databases to train their diffusion AI model (Stable Diffusion), without obtaining a license, breaching copyright. Images generated by Stability AI have subsequently closely resembled Getty's copyrighted images and, in some cases, even include a Getty Images watermark, which also breaches trade mark. Stability AI argues that the generated images are diffused from random "noise" and the user's prompt determines the output (i.e. the prompt asked for a Getty Images watermark). They claim that the generated images fall under "pastiche" (style imitating another work or artist), which does not infringe copyright. You can read more about the case on Gilbert + Tobin Law.

Several other content owners and artists have taken similar cases against GenAI companies.

Can AI-generated content be copyrighted?

Copyright laws vary from country to country. However, in most cases, copyright is granted to a human author/producer of the content. AI cannot claim copyright and you cannot claim copyright for something produced by GenAI. If you make significant edits to AI-generated content, you can claim copyright for your contributions but still not for what was produced by GenAI. For a more detailed example of copyright law related to GenAI, see the US Copyright Office's Copyright Registration Guidance: Works Containing AI-Generated Materials.

This may have implications for your use of GenAI. For example, if you are an artist and use GenAI in work you plan to exhibit or sell, you will not hold copyright over any part of it which was produced by GenAI, meaning others could use it without paying or acknowledging you. This is particularly problematic if the whole artwork is produced by GenAI. If you wrote a particularly detailed prompt to produce the image, you could potentially copyright that, but not the image itself. Similarly, if you are a researcher and use GenAI to produce images capturing your research for publication, you will have no rights to these images.

Reflection

Imagine that you have your own website where you regularly post articles, film reviews, and some reflective essays. You are using this as a portfolio of your work and your website is copyrighted. If this is publicly available, it is possible that it was scraped and used as training data for GenAI models without your permission. How do you feel about this? What if a GenAI tool outputs something similar to your work and it is used by someone else?

Imagine that you are trying to start a new business and use GenAI to create a logo for it. The logo looks nice so you use it for your branding. Who owns this logo? You, the AI company, no one? What if another company sees your business is doing well and starts to use this logo? Can they do this if it can't be copyrighted?

What should you do?

Consider copyright and intellectual property when interacting with GenAI tools. Ensure that you understand how GenAI models use data that you share with them. Do not submit copyrighted material or others' work into GenAI tools without permission. Be careful when inputting any of your own work/research as this may pose challenges to your intellectual property if the GenAI tool outputs something similar to your work for another user. Remember that if you use AI-generated images or other material in your work, you may not be able to copyright these.

Privacy concerns

GenAI raises significant privacy concerns, particularly regarding the collection and use of personal data in training models. Many AI systems are trained on vast amounts of publicly available and scraped data, often including sensitive or personally identifiable information without individuals' consent. This can lead to unintended privacy violations, such as AI models generating responses that inadvertently expose private details. Additionally, deepfake technology and AI-generated content can be misused to create realistic but false representations of individuals, leading to risks such as identity theft, misinformation, and reputational harm. The lack of transparency in how AI systems handle and store data further complicates efforts to ensure compliance with privacy laws like GDPR. As GenAI continues to advance, there is an urgent need for stronger data governance, user control over personal information, and ethical safeguards to prevent the misuse of AI-generated content.

Scams, fraud, and harmful content

GenAI is increasingly being used in scams and fraudulent activities, and to produce harmful content. LLMs are being used to make fraudulent websites and emails sound more convincing and credible. Deepfake videos, audio and images that impersonate individuals are also being AI-generated. There are several cases where deepfakes been used to persuade individuals to purchase something, vote a certain way or send money to fraudsters as in a recent case in Hong Kong where a finance worker paid out $25 million after participating in a deepfake video call with several individuals he thought were colleagues. Deepfakes are not only used for fraud, but also to create harmful and damaging content, such the AI-generated explicit images of Taylor Swift, and to spread disinformation. It is therefore important that we are vigilant about our own data privacy and in critically evaluating content that we engage with.

For an overview of AI-generated content being used for fraud, see this article from Reuters.

Reflection

You recently ran a survey and are going through the results. You think ChatGPT will be useful for organising these results more clearly so you plan to input them in and ask it to summarise and structure them. What should you check before doing this? What are the risks of this?

What should you do?

Check how any GenAI tool you interact with uses the data you input into it. Does this form part of its training data? Be careful not to input any private/personal information (your own or others') that you do not want shared. Always critically evaluate content you engage with and be wary of potential scams and disinformation.

The content in the GenAI Learning Hub is licensed under a Creative Commons BY-NC 4.0 license, except where otherwise noted. This license allows you to share or adapt the material for non-commercial purposes only, and with appropriate attribution to the original source.

Copyright © GenAI LEARN project team, University College Cork.