Yes. Lawsuits are being filed and the plaintiffs have every reason to expect success. Artificial intelligence (“AI”) is not really “intelligent” and it is not autonomous. Someone has created the software and programming — wrote the code — that instructs the AI to do what it does. That means that a person or business is legally responsible for what the AI does. This includes legal liability when an AI software program commits copyright infringement. Even if we ignore the question of whether an AI is “intelligent” and “autonomous,” an AI program is still owned by someone. The owner is responsible for the conduct of the AI program.
Take, for example, self-driving autonomous cars, trucks and vehicles. If a self-driving vehicle causes an accident, the owner of that vehicle is legally liable for any and all injuries and damage caused. The same legal principles apply when an AI software program commits copyright infringement. The legal concept is called “vicarious liability.” An employer, for example, is vicariously liable for the acts and omissions of its employees. In the same manner, an owner of an AI program will be vicariously liable for the acts and omissions of its AI program.
Admittedly, the legal theories are just now being tested. But, as stated above, copyright owners who have been the victim of AI copyright infringement have every reason to expect success with their copyright infringement claims.
What must be shown to prove copyright infringement?
At the federal level, copyright infringement is governed by the U.S. Copyright Act. Under the Copyright Act — 17 U.S.C. § 501(a)-(b) — there are only two legal elements that must be shown to prove copyright infringement:
- The person or business suing — the plaintiff — is the owner of a valid registered copyright; and
- The person/business being sued — the defendant — copied some amount of original expression from the copyrighted work
If an AI program is involved, then a third element will be needed: showing that the defendant was the owner, controller, or benefited from the AI program’s copying of the original expression/authorship protected by the valid registered copyright. Importantly, proof of intentional action is not an element. Thus, it is no legal defense to say “I didn’t know my AI was committing copyright infringement” or “it was an accidental infringement by the AI.” This makes common sense. To use the self-driving vehicle example again, there would be no legal defense to say: “I didn’t know my autonomous vehicle was going to cause an accident.”
What damages can be recovered by AI copyright infringement?
Under the Copyright Act, proof of copyright infringement allows the owner of the copyright to claim statutory damages or actual damages. Since actual damages are sometimes difficult to prove, many copyright infringement cases seek statutory damages. Statutory damages are awarded for each work of authorship that is infringed against and range from a low $750 per work up to $30,000. If infringement is proven to be intentional, then the statutory damages can be up to $150,000 per work.
Contact the Internet Law Attorneys at Revision Legal For more information, contact the experienced Copyright Lawyers at Revision Legal. You can contact us through the form on this page or call (855) 473-8474.
How AI Systems Actually Copy: Training Data and Output Reproduction
To understand how AI systems commit copyright infringement, it is necessary to understand, at a basic level, how generative AI systems work. Large language models and image generation models are trained on massive datasets of existing text, images, code, and other content scraped from the internet and other sources. During training, these models identify statistical patterns in the training data and use those patterns to generate new outputs. The training process itself involves reproducing and processing the copyrighted works in the training dataset — and several pending lawsuits argue that this reproduction, done without authorization and without compensation to copyright holders, itself constitutes copyright infringement.
Courts are currently divided on whether training-phase reproduction constitutes infringement or qualifies as fair use under 17 U.S.C. § 107. In Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015), the Second Circuit held that Google’s digital scanning and indexing of entire books for use in its book search function was fair use, at least in part because the resulting output was transformative and did not substitute for the original in the marketplace. AI companies have relied on this and similar decisions to argue that training-phase copying is similarly transformative. Copyright owners argue that AI training is categorically different: unlike a search index that displays snippets, AI systems generate outputs that can directly substitute for the original — and the training data is never deleted from the model’s “memory.”
Output Infringement: When AI Generates Content Too Close to the Original
Even if training-phase copying is ultimately found to be fair use, AI systems can commit copyright infringement through their outputs. When a generative AI system produces text, code, images, or music that is substantially similar to a specific copyrighted work in the training data, the output itself may infringe that work’s copyright. Courts apply the “substantial similarity” test to evaluate whether an allegedly infringing work copies protectable expression — not just facts, ideas, or style — from the original.
In the AI context, output infringement is most likely when the user’s prompt specifically asks the AI to reproduce, imitate, or build upon a particular copyrighted work, and when the AI’s training data included that work. Several reported cases have demonstrated that leading AI image generators will produce images closely resembling specific artists’ distinctive styles and specific copyrighted images when prompted to do so. Courts evaluating these cases will need to apply established copyright infringement analysis — substantial similarity, access, and copying of protectable expression — to AI outputs that are generated probabilistically rather than through deliberate human copying.
Who Is Liable When AI Infringes?
Multiple parties may bear liability for AI-generated copyright infringement, and the scope of each party’s liability depends on the facts. The developer or operator of the AI system — the company that built, trained, and deploys the model — faces primary liability under the vicarious liability and contributory liability theories discussed in the original post content. But users of AI systems may also face liability in certain circumstances.
A business that uses an AI system to generate content for commercial purposes — marketing copy, product descriptions, website content, software code — and publishes that content without any review for potential copyright infringement, takes on liability risk. If the published content substantially reproduces a copyrighted work from the AI’s training data, the business has published infringing material, even if the AI generated it automatically. The absence of intentional copying is not a complete defense to copyright infringement; copyright is a strict liability regime in which intent to infringe is not required, though it affects the amount of statutory damages available.
Extra, Extra!
Related Posts
Artificial intelligence has become part of nearly every business operation. Businesses now use AI tools to write marketing copy, generate product images, compose emails, draft social media posts, and produce video and audio content at a scale that was not possible a few years ago. The efficiency gains are real. But so are the legal […]
Read more about The Risks of Using AI-Generated Content in Your Business →
Put Revision Legal on your side