Legal Considerations of Artificial Intelligence for Student Journalists

Last updated August 2024

Since our founding in 1974, the Student Press Law Center has assisted student journalists with many novel issues of media law, most arising from new technology. The arrival of Artificial Intelligence (AI) brings the latest concerns to our doorstep, as many students and advisers have directed questions about the already pervasive technology to our Legal Hotline. 

This guide is intended to give student journalists and educators a basic legal framework through which to evaluate the most common questions we are being asked as the technology continues to evolve. It is not a primer on how AI platforms work, nor does it address other ethical, editorial and educational ramifications of AI. Please see the end of this guide for other resources on those perspectives.

Ultimately, this is new to all of us — including courts, lawmakers and lawyers. The answers provided in this guide will almost certainly change as the technology and the law evolves. But it’s our best effort — right now — to offer guidance on the novel legal issues created by this extraordinary, and  imperfect, new tool.

• Introduction: The AI Genie Is Not Going Back in the Bottle
• Bottom Line: You Are Responsible For What You Publish or Broadcast
• Libel and AI
• Copyright and AI
• Privacy and AI
• Other Helpful Resources

Introduction: The AI Genie Is Not Going Back in the Bottle

Many school and college administrations around the country are considering or have already “banned” the use of AI for class work, including for publications. This is an unrealistic prohibition, however, given how pervasive and accessible these platforms have become. The 1990s saw similarly widespread rules and frequent outright bans on student use of the Internet, but such prohibitions now just seem silly — and a serious missed opportunity to educate. Student journalists, like all students, benefit when they are trained and prepared for the world they will actually inhabit. We encourage student media programs to consider — and we will work with those interested in creating — thoughtful policies that embrace (or at least acknowledge) this latest technology instead of eschewing it.

To educators: We encourage you to assume your students are already using AI tools and to experiment with it yourselves. For example, there may be no better way to dissuade careless AI use than to give your own anecdotes as to how you “stumped” the platforms or caught them making errors. Give ChatGPT, Google Gemini, Jasper Copilot or one of the many new AI platforms a try. Play around. Try completing some of the assignments you give in class using various platforms, and note the result.

Bottom Line: You Are Responsible For What You Publish or Broadcast

The law is clear that we are always responsible, or liable, for what we personally say or create and publish. Some new journalists are surprised to learn, however, that they can also be held responsible for republishing things that others — the law calls them “third parties” — say or create. In much the same way that we use third-party content like guest columns, wire services or letters to the editor, AI functions as another third-party contributor. And in all these situations, the basic third-party liability rule applies: If you re-publish or re-broadcast someone else’s words or work, you are responsible for it. This third-party liability can cause problems in three of the substantive areas discussed in this guide: libel, copyright and privacy.

Libel and AI

Let’s cut to the chase: If you repost AI-generated material that contains libelous content, you can be on the hook for it. You must fact-check everything provided by AI, just as you should anything else.

As a refresher, libel is the publication of a false statement of fact that seriously harms someone’s reputation. A statement can only be defamatory where it is presented as a “factually true” assertion (not a statement of opinion). This is what AI does. AI processes vast amounts of information from existing sources that “train” the model to better hone their output. When prompted with a question, AI inherently avoids expressing possibilities; instead, it is designed to generate definitive, factual (or seemingly factual) statements. 

But AI is far from perfect and factual errors are common. Mistakes made by AI platforms are known as “hallucinations,” and they are often easy to spot when playing around with AI platforms. Some hallucinations are more subtle, such as when AI presents facts from a publication but inaccurately attributes them. Similarly, AI may make up information or quotes that seem plausible, which one Wyoming reporter learned the hard way. (“Those are words that I very well could have said, but they are not what I said,” said one source that AI allegedly misquoted.)

Sometimes the mistake is relatively “painless.” For example, after training on an article that accurately covered a lecture the person gave at a certain university, AI mistakenly interprets the information and creates the hallucination that the individual graduated from that university instead. The information is false, but it’s not likely to seriously damage the person’s reputation.

Unfortunately, some mistakes can be much more significant and result in serious harm to a person’s reputation that could serve as the basis of a lawsuit. For example, if AI absorbs an article from a reliable source about a criminal lawyer that discusses some of the cases he took on, but a resulting hallucination falsely reports that he was convicted of the crimes mentioned in the article, his reputation could be seriously harmed. 

It may seem obvious, but you should carefully fact-check all AI output. While AI offers the opportunity to “check itself,” don’t let that give you false confidence to breeze past your own fact-checking. So far, AI’s ability to explain its own process of delivering information is limited. Nothing, at present, substitutes for a human fact-checker. After all, when we asked ChatGPT in August 2024 about the sources of its information, the platform indicated its last “knowledge update” was October 2023!

Playing Out in the Courts

Courts are slowly tackling some of the new legal issues raised in libel cases involving AI. For example, in Georgia, a talk radio host complained that OpenAI libeled him, and he sued the company whose employees designed the software. (Walters v OpenAI, LLC). In this case, OpenAI allegedly “hallucinated” that the host “misappropriated funds for personal expenses without authorization or reimbursement, manipulated financial records and bank statements to conceal his activities, and failed to provide accurate and timely financial reports” for an organization he never even worked for. In fact, the host had won an award from the organization. 

The host claimed the false information generated by OpenAI seriously harmed his reputation. He argues that, because OpenAI was aware of the possibility of hallucinations, they should be liable for such mistakes when they cause harm. In its defense, OpenAI argues that it “has been consistently transparent about the limitations and responsible use of this emerging technology.” As the company told the court:

“AI-generated content is probabilistic and not always factual, and there is near universal consensus that responsible use of AI includes fact-checking prompted outputs before using or sharing them. OpenAI clearly and consistently conveys these limitations to its users. Immediately below the text box where users enter prompts, OpenAI warns: ‘ChatGPT may produce inaccurate information about people, places, or facts.’ Before using ChatGPT, users agree that ChatGPT is a tool to generate ‘draft language,’ and that they must verify, revise, and ‘take ultimate responsibility for the content being published.’”

Only time will tell whether the court agrees with the host’s libel claim or OpenAI’s defense. Regardless, it’s still a safe assumption that even if OpenAI is ultimately not held liable for what it produces, you would be responsible for re-publishing any of its false and defamatory information as fact. We should take notice of the warning OpenAI gives us: This technology is not reliable right now, and it should not be used as a substitute for traditional and responsible newsgathering.

Recommendations

• Don’t use AI if you wait until the last minute to draft or edit a story.
• Fact-check information that can be true or false (remember: output is not libelous if it is pure opinion).
• If there is a situation where AI must be used, such as in covering the output of AI itself, it may be helpful to include an explicit disclaimer indicating that AI’s output is not always factual (much like we must make parody or satire obvious to the reader).
• Pay attention to the websites you use as starting points for your newsgathering: more unscrupulous sites are using AI for quick and cheap information to generate advertising revenue.

Copyright and AI    

Perhaps more than any other issue, copyright law and AI are — at least for the moment — on something of a collision course. Copyright protects the works of creators. AI uses the works of those creators to do what it does. While it may seem that AI is creating wholly original text and images when it responds to user prompts, it’s not. AI in 2024 pulls facts and materials from other creators — for now, mostly human beings — to craft its responses, usually without permission. That raises important questions about the legal rights of creators and the legal risks for AI users.

For background on copyright, the Student Press Law Center has several resources explaining the law and how it interacts with student media. In general, you own the copyright to material you create from the moment it’s created. Copyright law then protects you, in most circumstances, against the unauthorized use of that material, both in its original form and even if modified by others (the law calls these “derivative works”) unless the modifications are so substantial that they effectively create a new work that stands on its own. These copyright protections apply whether the material at issue is a story, photo, video, etc. In the same way, copyright generally prohibits you from using materials owned by others without their permission. 

But what if it’s AI — and not another human — that uses copyrighted materials without permission? For example, what if AI poaches your photo of the big game last weekend and uses it on a website entirely generated by AI? What if it also rewrites your story — sort of — about the big game and publishes it? Who, if anyone, is responsible? The creators of the AI platform? The owners of the website that re-published the photo or story? Unfortunately, as with the above libel discussion, we are still in the early stages of courts answering these questions.

Uncertain Ownership

Remember that AI platforms amass vast quantities of information to “train” themselves. This training can come in the form of text or images, both protected by copyright, often without seeking permission. This has proven controversial, including within the news media, with The New York Times suing OpenAI and Microsoft for using its content without permission, while the Associated Press has signed a deal with OpenAI to make AP’s news archive available for AI training.

In several ongoing lawsuits against AI platforms, creators allege that — because AI trained on their protected work — any output is a derivative work that also would be protected by their original copyright. If this legal theory proves true, it is conceivable that anyone who later publishes or broadcasts that output — including student journalists — might also be in violation of the original creators’ copyright. At least a couple of judges have stopped short of making such a ruling after determining the AI output was not “substantially similar” to the original protected material. It is too early to be sure if a “substantial similarity” test is where the law will end up. (Similarly, one could also argue that AI output may constitute enough of a “transformative use” that it could fall under the “fair use” exception to copyright, but that is also unsettled.)

You may have heard that, because of this uncertainty, some platforms have offered legal guarantees built into their AI-heavy products. For example, Adobe has developed a standalone generative AI web application called Firefly, and boasts covering your legal fees related to any potential copyright infringement by its output. Microsoft Copilot offers similar protection. It may be that other AI generative platforms will follow their lead.

If the AI output is not owned by the input’s creators, that leaves a few other options for who is the owner: (1) the AI company, (2) the user creating the prompt or (3) no one. At least one AI platform — ChatGPT — says it’s not them. As its “bot” will tell you if you ask for ownership information: 

“If you provide the input and prompts for the narrative I compose, you would likely be considered the author and owner of the resulting content. This is because you initiated the creative process and provided the direction for the narrative.” 

However, the U.S. Copyright Office chose option No. 3, rejecting any protection of fully AI-generated images, reminding us that “copyright can protect only material that is the product of human creativity.” The U.S. District Court for the District of Columbia agreed, noting that copyright doesn’t “protect works generated by new forms of technology operating absent any guiding human hand…” But if 100% machine-created material is not protected, what about 90%? 50%?

Given the uncertainties about copyright ownership right now, student media that want to protect themselves should consider either (1) avoiding directly publishing or broadcasting AI output entirely, or (2) creating procedures to make sure that any published output is not substantially similar to other works. Using tools that guarantee legal fees may be one way to assuage those concerns.

AI-Generated Websites Using Content From Student Journalists

We’ve heard from an increasing number of student journalists through our Legal Hotline recently about their materials being poached by websites powered by AI, often word-for-word and image-for-image. In fact, one company that manages AI-populated news sites even contacted a Pennsylvania student editor to request that the high school newsroom change their online headlines to be more easily picked up by their AI, evidently unconcerned about the students’ copyright. 

Whether these uses constitute a copyright violation will generally depend largely on how closely they copy your materials. Copyright law does not protect facts or ideas themselves, so if AI substantially rewrites your story using the facts you reported, there may not be much that can be done legally. (Of course, that doesn’t mean you can’t reach out to them to request they stop doing so on ethical grounds.)

Copyright law does, however, protect the expression of those facts — in other words, the exact phrasing and arrangement of a story. So, if the AI-powered website reposts your story word-for-word — or close to it — that is a copyright violation. When this happens, it is essential to reach out to the infringer as soon as possible.

If it is a first-time offense that you suspect was out of simple ignorance of your copyright, find the website’s contact information to alert the infringer of your copyright and request that it be removed immediately. 

If this does not work, or you suspect that the infringement was deliberate, a cease-and-desist letter may be appropriate. In such a letter you will want to claim ownership over your work and describe it, indicating that their use violates your copyright. There are a variety of remedies you can request: (1) require that your material be properly attributed; (2) ask for an apology; (3) demand that the material be removed immediately or that the entity cease financially benefiting from your work; and/or (4) show specific monetary damages from the infringement, and demand that you be paid. (If you have officially registered your work with the U.S. Copyright Office, you may have additional remedies available.) This letter can be sent via certified mail through your local post office at low cost, which serves as evidence that the poacher received notification. 

If the website originates from within the United States, you also may explore the Digital Millennium Copyright Act (DMCA) after sending the letter. Unfortunately, SPLC is seeing an increasing number of offshore AI-generated sites unlawfully using student media material. The owners of such sites can be difficult to trace and even more difficult to hold responsible.  If you have difficulty with these processes, as always, please contact the Student Press Law Center’s Legal Hotline.

Recommendations

• Be vigilant about your own reporting and photography property: look for your work being poached by other entities.
• Establish a policy for your newsroom on when and how AI can be used, and — if that includes directly publishing AI output — how you will ensure it doesn’t include copyrighted material.
• Make sure to do your homework if you plan to publish a composite image created by AI. If that image is not transformative enough from an original work, you may be on the hook for copyright infringement.
• Look for software that comes with “legal protections” advertised like Adobe Firefly or Microsoft Copilot.

Privacy and AI 

Invasion of privacy is currently the least-litigated of the topics covered in this guide, but it has no less gravity if we are not careful when using AI. Simply put, it is increasingly challenging to keep our information online confidential or anonymous, and as a result, AI may scoop up otherwise private information while it trains. This raises questions of liability for those who publish private information included in AI output, but it also implicates more general concerns that journalists’ confidential information may be accessible and shared as well.

As a baseline reminder, there are traditionally four types of invasion of privacy:

1. Public disclosure of private and embarrassing facts
2. Intrusion
3. False light
4. Misappropriation

Each type is described in greater detail in SPLC’s privacy law guide. The first type — publishing highly private, intimate details about a person — is the one most likely to cause a problem in the context of AI, and we’ll focus on that one here. Unlike libel, where truth is an absolute defense, something could be absolutely true but still qualify as an unlawful disclosure of private information. However, a defense of newsworthiness might apply. The balance is between the privacy of the individual claiming a violation and the importance of the information to public interest or concern. If the information is a matter of significant public interest or concern, its publication is newsworthy and cannot be a public disclosure of private facts no matter how embarrassing it is. Courts have found that almost any information about a well-known public figure or public official will be considered newsworthy.

Keep in mind that AI training materials may come from anywhere on the Internet — including data leaks from large companies that store sensitive information, such as very personal financial or health details. As new lawsuits argue, in addition to these data breach results, other sensitive information swept up by AI may include private conversations and information about minors. An AI platform, as a single “mind” of information, can be asked about any sensitive information it has processed during training. The platform cannot effectively distinguish between sensitive and non-sensitive items. Therefore, even if some very personal information produced by an AI-powered search might be correct, it could be an invasion of privacy to publish something that had mistakenly or unlawfully (e.g., a “hack”) been made public in the first place.

Since the advent of AI, privacy experts have been scrambling to keep data secured (or “anonymized”) from the probing arm of AI’s training, but it has been difficult. The work of anonymizing the data gets even harder as AI systems become more sophisticated.

If something seems a bit too personal online, it may have been the result of the above process.

Student journalists, in large part, must rely on a sense of what seems like an “overshare” when researching online information about an individual. Some information that we take for granted (like the granular detail we sometimes add to public websites like LinkedIn) can be combined to identify other data about us. Such other data could include information we would have preferred not to be public. The result: AI can assemble startlingly accurate composite profiles of individuals and share these details in its output without being explicitly asked to do so. Alternatively, you may have relied on AI-assembled articles that include very private information. Diligent investigation is paramount here.

Examples of sensitive information that may be provided by AI

• Data leaks: customer or patient names and their addresses, phone numbers, dates of birth, spending habits, social security numbers, medical information
• Social media: information in an account ordinarily only accessible by consent (e.g., friending someone on Facebook to see their full profile) or content behind a paywall
• News media: identity of confidential sources of journalists or other unpublished materials that might be accessible via a hack or data breach.

Recommendations

• Get into the habit of asking the AI platforms for the sources of their information, and ask for details.
• When researching a story, if you find what appears to be sensitive information generated through AI platforms, you may need to do some further digging as to the information’s origin.

Other Helpful Resources

• The National Scholastic Press Association’s Code of Ethics for high school journalists now includes a section on Generative AI (Part 23). It is free for members or available for purchase.
• The Journalism Education Association’s Digital Media Committee offers a sample editorial policy and tips for creating a policy.
• The Poynter Institute provides a starter kit for creating a newsroom AI policy.
• The Journalistic Learning Initiative has created Murrow, an AI chatbot that coaches students in journalistic story development and writing.