What marketers (and others) should know about the legal landmines in AI content creation

The biography for my co-author, Eric Lambert, is at the bottom of this article.

Note: This article is based on information available as of the date this article was published, April 15, 2024. As developments in the AI space are fluid, information may no longer be up to date depending on when you’re reading this article.

Key takeaways:

• AI output is a prediction based on a combination of the model’s training data and the human’s input.

• Content or information you submit to an AI platform may be used to generate copy and other content for others.

• In most cases, AI tools say you own the copyright of AI output, to the extent they can give it to you (see below), but if others’ copyrighted material was used to generate the output, your ownership may be “polluted.”

• You’re responsible for inaccuracies or copyright infringement in AI-generated content.

• US law does not allow for copyright protection on works created solely by AI. UK law provides copyright protection for AI-generated works with no human creator.

• Stay on top of the rapidly changing landscape of AI laws, regulations, and best practices.

Many marketing professionals are increasingly using AI in content creation, as it enables the generation of everything from catchy social posts and website graphics to comprehensive blogs, white papers, and videos. However, the proliferation of AI use in the content generation process raises important legal questions and considerations, including: 1) How does an AI platform use data and information provided to it for its own purposes? 2) Who “owns” the copyright to AI-generated content? and 3) Who bears responsibility when the published output is incorrect, made up, or plagiarized?

These questions aren't hypothetical; they carry real-world implications for anyone using AI. For marketers, understanding how AI tools work and the legal considerations and pitfalls is crucial—not only to safeguard your work but also to navigate the challenges of this new territory. We're at a juncture where laws are beginning to catch up with technology, making staying informed essential for responsible AI use.

A quick overview of Generative AI

At the heart of generative AI is a Large Language Model (LLM). An LLM takes an input, such as a natural language question, for example, “What color is the sky?”, and uses a type of neural network called a “transformer model” to generate an output in response, such as “the color of the sky is typically blue...” An LLM’s neural network is trained to recognize context by understanding associations between content elements such as each of the words in this sentence and predicting what word comes next. AI tools that generate images or video from text, such as DALL-E, Midjourney, or Runway, use both an LLM and another neural network called a “diffusion model” to generate images, animation, and video again, by predicting what comes next.

Understanding context is critical for AI’s ability to interpret natural human language and determine its meaning. The larger the data set used to train the model, the better the model generally is at determining context, correctly predicting what comes next, and providing relevant, accurate answers. In fact, some LLMs such as ChatGPT, Gemini, and Claude are trained on a good portion of the data available on the Internet, which gives it its broad knowledge, but is also why it can regurgitate, or copy, information it was trained on in its output, or provide incorrect answers (known as “hallucinating”) by responding with made-up or fake information.

The key thing to remember here, especially as it pertains to legal landmines, is that when using AI, the output is a prediction based on a combination of the model’s training data and human input. Different models and different inputs will elicit different outputs—some of which are better, more or less unique, and more factual than others. 

How does an AI platform use data and information provided to it for its own purposes?

The terms of use or license agreement for large, public AI platforms often state that they have the right to use the content of questions submitted to the AI tool, and the output generated by that tool, to further train their models. In other words, any content or information you submit to an AI platform may be used by the AI tool to generate copy and other content for others. For example, ChatGPT’s Terms of Use from January 2024 are very clear that it can use both ChatGPT inputs and outputs “to provide, maintain, develop, and improve our Services….” Midjourney’s March 2024 Terms of Service state that the user gives them a perpetual license “to reproduce, prepare derivative works of, publicly display, publicly perform, sublicense, and distribute text and image prompts You input into the Services, as well as any Assets produced by You through the Service.” 

For example, if you provide an AI tool with text input to generate ideas, copy, images, or other content for use in a campaign, or if you use an AI tool to create a derivative work of your existing content or images (e.g., you want Gemini to create a blog about an existing white paper or Adobe Firefly to create a similar image to an existing campaign asset), both the input and content (text and images) you put into the tool, and the output you get back, may be used to train the LLM and could show up in other users’ results. This is why you must be very careful in what you provide to an AI tool as you may be inadvertently taking confidential information or your proprietary content and making it publicly available through the AI tool for others looking for similar ideas or inspiration.

Who “owns” the copyright to AI-generated content?

Generative AI tools such as ChatGPT, Gemini, and Midjourney are all generally clear that “as between you and the generative AI tool”, you own the output you generate from the AI tool to the extent permitted by applicable law. This means that the AI tools are “passing the buck” to the user with respect to ownership issues under an evolving legal landscape. Let’s dig into the two things here that can trip you up and get you in trouble legally:

• “As between you and the AI tool.” This is saying that when it comes to questions of who owns the output—you or the AI tool, such as ChatGPT—the answer is you. However, there’s a huge disclaimer here: if the LLM content used to generate your output included content owned by someone else (e.g., they have the copyright to that content), you may not truly or fully own that output. We’ll talk more about this later.

• “To the extent permitted by applicable law.” The U.S. Copyright Office states that copyright law protects original works of authorship, including various forms of artistic and literary works. However, the ownership question becomes complex when AI or a combination of AI and human effort produces these works. In the United States, the current position of the U.S. Copyright Office is that the author of a work of authorship must be human, not an AI. For example, the output generated from a text prompt such as “create an image of a toaster with the face and hands of a hamster eating a toaster pastry with strawberry frosting” will most likely not be copyrightable. This is consistent with a federal district court decision in August 2023 that ruled in favor of the Copyright Office with respect to its decision to deny a copyright for an autonomously-generated AI image. However, it’s not yet clear how much human effort and involvement is required to allow for copyright to apply. That creates a legal gray area for content written by a human and enhanced or augmented by AI, or AI-generated content based on detailed human outlines or creative prompting.

One important note regarding creating derivative works of your own content. If you prompt an AI to create a blog based on a white paper you wrote (and thus own the copyright), that blog is considered a derivative of the white paper and you would own the copyright. However, things can get cloudy quickly. For example, if a human creates a derivative work of an image in which that human owns the copyright, that person owns the copyright to the derivative work too. However, if you prompt an AI tool to take your image and create a new image in its likeness, there’s a chance that the AI tool could use elements from other pictures or sources to generate the derivative work. If that happens, it could potentially “pollute” your copyright by adding another person’s content to the derivative work. It may be a small risk, but it’s there. As long as you or the AI doesn’t add any third-party copyrighted content, you should own the copyright in your derivative work to the same extent you own the copyright in the original image. 

Across the pond, the United Kingdom takes a different approach, offering copyright protection to AI-generated works without a human author, instead assigning copyright to the person who prompted the AI. This contrast highlights the complexity and inconsistency of international copyright laws regarding AI. The UK is currently developing a copyright and AI “code of practice” which is anticipated to be released later this year and is intended to balance the use of copyrighted works by AI developers against content creators’ rights in those works.

For marketers, this maze of legal considerations means understanding the extent of AI assistance in your content creation and the copyright laws applicable in your content's jurisdiction is critical. The situation becomes even more complex on a global scale, where rights to AI-generated content can vary significantly between countries.

Who is responsible for inaccuracies or copyright infringement in AI-generated content?

As marketers, we rely on the accuracy and originality of our content to uphold our brands' reputations and engage our audiences. So, when AI-generated content is used, addressing liability for mistakes or copyright infringement becomes crucial. To establish copyright infringement, a copyright holder must prove that the infringer copied the copyrighted work and that the AI-generated content is “substantially similar” to the copyrighted work. The “copying” element may be provable using circumstantial evidence that the infringing party had access to the work, e.g., that the AI tool that generated the AI content was trained on the copyrighted work. While some (AI tool owners, for example) believe that it’s very unlikely that an AI tool will create an infringing work, others (Getty Images and the New York Times, for example) feel that the likelihood of copying is far greater.

Like human-authored content, the publisher bears responsibility for any factual errors or unintended plagiarism, leading to legal and reputational consequences. Just look at what happened to Air Canada when their chatbot gave inaccurate information to a customer. AI could also conceivably create content that infringes other types of intellectual property, such as trademarks (a word, phrase, or slogan that is used to identify a company or its products or services and differentiate them from other companies, products, and services) and trade dress (a form of trademark that relates to visual appearance, such as the shape or design of a product or its packaging).

Depending on the type of license under which you use an AI tool (individual vs. business/enterprise), you, and not the AI provider, may be responsible for any copyright or other intellectual property infringement in AI-generated content. While some AI tools may defend and indemnify certain clients (e.g., those on business or enterprise versions of their tools) against settlements and judgments of copyright claims arising from output generated from their service, which is good for content creators, your company will still have to deal with the lawsuit and the associated time and energy (e.g., document discovery, depositions, etc.) as well as reputational hit that comes from those searching for lawsuits involving your company finding it. 

Additionally, the AI platform agreement may have restrictions that limit the scope of this protection, such as a limitation of liability. These AI platforms may give you some financial protections, but won’t fully insulate you. If you’re using the version of an AI tool designed for individuals (even as a business), there are likely no indemnity protections from copyright or other intellectual property infringement.

This level of responsibility underscores the necessity of rigorous checks and balances in your marketing reviews. Remember that AI can make mistakes, and in some instances may “hallucinate” and just make things up, or may create content using existing works on which it was trained. Using AI doesn't absolve us of the duty to verify information’s accuracy and ensure content is free from plagiarism. In fact, it adds a new layer of due diligence—you must substantiate and verify every claim when using AI-generated content to ensure it is factual and is unique in its presentation. This can involve a mix of fact-checking, sourcing credible information, and using tools, such as Copyscape, Plagiarism Checker, and Originality.ai, designed to detect similarities with existing content. Done well, these extra checks help you avoid legal pitfalls and maintain the trust of your audience and the integrity of your brand. 

Adapting to emerging legal standards and ethical practices

Industries and businesses are working to figure out how existing laws (e.g., marketing laws, privacy laws, and copyright law) apply to generative AI tools and use. Countries around the world, from Argentina to the United Arab Emirates, are implementing AI laws and/or policies. In the US, as AI technology becomes more prevalent in marketing, state legislators are also beginning to pass legislation putting rules, obligations, and limits around the use of generative AI tools. A notable example comes from Utah, where a recent law mandates the disclosure of AI-generated content. Specifically, the legislation requires that any person using AI to create or distribute content must disclose AI's involvement. This kind of legal development is significant—it's an early instance of state-level regulation in the United States that addresses the legal accountability and ethical use of AI in content creation, and it may pave the way for similar laws in other states. 

It’s almost always the case that the pace of legislation is far behind the pace of innovation, and that’s certainly true for AI. Legislators are scrambling to catch up with the pace of development in generative AI tools. From a compliance perspective, consider the “herd on the savannah” model—the safest place is generally in the middle of the herd, meaning if you are using reasonable efforts to comply with the developing AI laws, rules, and regulations and are applying best practices in your use of AI tools, you’re generally less likely to find yourself in trouble than someone that’s at the front of the herd (they want to be cutting edge) or at the back (they’re not paying attention to the developing legal and ethical landscape).

Emerging laws like those in Utah, combined with the lack of clear and mature guidance on copyright protection for AI-supported works, means that as marketers, we need to stay informed and be ready to adapt our practices as new regulations emerge, i.e., to “move with the herd.” For starters, marketing must include clear statements regarding the role of AI in their creative output, using frameworks such as the AI Contribution Scale, to remain compliant and enhance consumer trust by openly acknowledging the use of AI tools. 

Beyond that, we are responsible for using AI tools in ways that are fair, unbiased, and respectful of privacy. This includes being mindful of the training data AI, ensuring it doesn't perpetuate existing biases or invade personal privacy. Upholding these ethical standards is not just about avoiding legal repercussions; it's about building a marketing function that’s inclusive, diverse, and trustworthy.

Staying informed and proactive with AI-generated content

AI-generated content opens up thrilling opportunities while introducing new legal and ethical challenges for marketers and others. We must actively engage in understanding copyright laws, complying with new regulations, and upholding ethical standards. Together, we can confidently navigate this evolving landscape by staying informed, documenting AI involvement, ensuring the originality of our content, and maintaining transparency with our audiences.

Important note: We wrote this article to give marketers an intro-level understanding of the legal challenges of using AI in their content creation. This article is not intended to be and should not be relied on as legal advice. Make sure to consult a legal professional familiar with AI and copyright law if you have any questions or to evaluate your own particular situation. 

AI disclosure: The hero graphic was AI-directed using DALL-E on April 8, 2024. I, Natalie Lambert, contributed as the human element in this activity.