AI: Artificial Intelligence: Copyright and Intellectual Property

Artificial intelligence offers several challenges related to copyright and intellectual property, including those related to: 

• Training data: Can copyrighted materials be used to train for-profit AI tools? Generative AI tools are generally trained on extensive and broad sets of training data that could include images, text, video, audio, raw data, and other information types. AI developers such as OpenAI and Google have not always been transparent about whether they have trained their AI on copyrighted material, and many copyright holders have filed lawsuits against AI developers, claiming their copyright-protected works have been improperly used to train AI tools. 
• Information uploaded or added to an AI tool: When you upload content to an Ai tool, are you handing over copyright ownership to the developer? The answer depends on the fine print that you agreed to when you uploaded or entered your content to the AI tool. Similarly, you might be violating copyright or licensing agreements if you upload copyrighted material to an AI tool, such as uploading a pdf of a journal article to an AI tool that summarizes academic articles. Questions about uploading library-licensed content to an AI tool should be directed to your librarians. 
• AI output: Can you copyright content fully generated by an AI tool? This issue is still being decided in courtrooms around the world, and it's still unclear what the future holds on this topic, as copyright is generally granted only for human-created content. If you do apply for a copyright with the US Copyright Office for content that was partially created by AI, you must include a statement attesting to your use of AI: “An applicant who creatively arranges the human and non-human content within a work should fill out the 'Author Created' field to claim: 'Selection, coordination, and arrangement of [describe human- authored content] created by the author and [describe AI content] generated by artificial intelligence.”

If you have any questions about artificial intelligence and copyright or other forms of intellectual property, please visit our copyright guide.

If you are planning to publish your own work and are considering using AI tools at any part of the research or writing process, be sure to check out the publishers' policies related to artificial intelligence:

• Elsevier AI author policy
• Taylor & Francis Clarifies the Responsible use of AI Tools in Academic Content Creation 
• IEEE submission policies
• Nature Statement on authorship and AI
• PLOS ONE
• Sage Assistive and Generative AI Guidelines for Authors
• Wiley - Best Practice Guidelines on Research Integrity and Publishing Ethics
• Oxford Academic - Instructions to Authors
• Emerald Publishing’s Stance on AI Tools and Authorship

Publishers are also developing policies related to the use of artificial intelligence for the peer-review process, offering guidance on whether peer reviewers can use AI tools as part of the review process. 

Remember: Text, images, and other content you add to generative AI tools may be stored and used to train newer iterations of AI Assistants and Tools.  If you are sharing information that is confidential, under copyright, or that you may not want to redistributed this can be a major concern. If you want to protect you work, data, and intellectual property, taking the following steps is recommended: 

• Read the website’s terms of service to see if the site permits data mining
• Avoid putting your work into a generative AI tool with unclear or lax data privacy policies
• Consider adding a robot.txt file to limit web crawlers from accessing your site
  Note: This may also affect the way that your website appears in search results

Text Protection

• Writer, programmer, and tech consultant Alex Reisner developed a searchable database of books included in Books3, which is at the heart of the Authors Guild v. OpenAI court case. 
• Changing characters in works, like replacing the letter “a” with an “а” from the Cyrillic alphabet will make it more difficult for language models to be trained on your work, but it will also render the text unreadable for anyone using a screen reader. 

Image Protection

• “Have I been trained?” can help you assess if your work has been used to train AI models. 

• Consider using Glaze or WebGlaze to disrupt the AI training process.

• Nightshade not only protects your art but also poisons AI models.

Audio Protection

• AntiFake is a python program that introduces noise into audio files to prevent deepfakes. 
• The Federal Tade Commission (FTC) has been monitoring the impact of voice cloning on news agencies and the entertainment industry. In November 2023, they hosted a challenge to help detect and evaluate cloned voices.