Copyright in the AI Era: Legislating Against Digital Deepfakes

In an era where digital reality is increasingly malleable, the rise of artificial intelligence has introduced a powerful and perilous new tool: the deepfake. Combining the terms "deep learning" and "fake," deepfakes are synthetic media in which a person’s likeness is replaced or manipulated with astonishing realism. This technology, capable of creating convincing but entirely fabricated videos and audio clips, has exploded from a niche curiosity into a mainstream phenomenon with profound societal implications. While offering creative potential in fields like entertainment and art, deepfakes also present a formidable threat, enabling the spread of misinformation, the creation of non-consensual pornography, and sophisticated financial fraud.

At the heart of this new challenge lies a fundamental conflict with long-established legal principles, particularly copyright law. The very essence of deepfake creation—the unauthorized copying and manipulation of voices, faces, and performances—collides with the core tenets of intellectual property. Yet, the legal frameworks designed in a pre-AI world are struggling to keep pace, leaving a trail of complex questions about authorship, infringement, and liability. As technology continues to outrun regulation, lawmakers, courts, and tech companies across the globe are scrambling to erect legal guardrails against a technology that can convincingly blur the line between truth and fiction.

The Core Copyright Conundrum: An Old Law in a New World

Traditional copyright law, a cornerstone of protection for creators, has proven ill-equipped to handle the unique challenges posed by deepfakes. The legal doctrines that have governed creative works for centuries are now being tested by a technology that operates in a legal gray area, challenging the very definitions of authorship, originality, and infringement.

The "Human Authorship" Hurdle

In many legal systems, most notably in the United States, copyright protection is fundamentally tied to human creativity. The U.S. Copyright Act of 1976 specifies that for a work to be copyrightable, it must have a human "author". This principle has been repeatedly affirmed by the U.S. Copyright Office and federal courts, which have maintained that works created solely by machines or artificial intelligence cannot be copyrighted. The reasoning is that copyright law is intended to incentivize human creativity by granting exclusive rights, an incentive that non-human entities do not need.

This "human authorship" requirement creates a significant paradox. The output of a generative AI, even if prompted by a human, is largely determined by the machine's internal processes—a "black box" that even its creators may not fully understand. The Copyright Office has argued that merely providing a text prompt to an AI system does not meet the "traditional elements of authorship," as the AI itself is executing the creative expression. Consequently, most AI-generated content, including deepfakes, is not protected by copyright. This leaves a void where neither the person whose likeness is used nor the creator of the deepfake can easily claim copyright ownership over the final product.

Derivative Works vs. New Creations: A Murky Distinction

Copyright law grants original creators the exclusive right to produce "derivative works"—new creations based on one or more preexisting works, such as a movie adaptation of a book. A key question is whether a deepfake should be considered a derivative work of the original footage, images, or audio recordings used to create it. If so, its creation without permission would constitute copyright infringement.

However, this is not a simple determination. The creation of a deepfake often involves compiling and learning from countless data points, making it difficult to prove that a "substantial part" of any single copyrighted work was used in the final output. Furthermore, some argue that if the deepfake is created for purposes of parody, satire, or commentary, it may be protected under the "fair use" doctrine, which permits the limited use of copyrighted material without permission from the copyright holder. The line between an infringing derivative work and a transformative new piece protected by fair use is often blurry and subject to intense legal debate.

The Inadequacy of Existing Frameworks

Classical copyright law was designed to protect original works fixed in a tangible medium, but it offers little direct protection for the intangible elements that deepfakes exploit, such as a person's voice, facial features, or mannerisms. While a specific photograph or film is copyrighted, the subject of that photo or film rarely owns the copyright themselves; it is typically held by the photographer or the film's producer. This leaves individuals, particularly non-celebrities, with limited recourse under copyright law when their likeness is stolen and repurposed in a deepfake. The U.S. Copyright Office itself acknowledged in a 2024 report that current laws are inadequate to address the risks posed by deepfakes and recommended that Congress consider specific federal legislation.

A Patchwork of Protections: Life Beyond Copyright

Given the limitations of copyright law, victims of malicious deepfakes have had to turn to a collection of other legal theories to seek justice. This "patchwork" of protections, while useful, often provides incomplete solutions and varies significantly between jurisdictions.

Right of Publicity: This legal principle protects an individual's right to control the commercial use of their name, image, and likeness. If a deepfake uses someone's identity to endorse a product or service without permission, it could be a clear violation of their right of publicity. However, these laws are not uniform; they vary from state to state in the U.S., with some states offering robust protection while others have limited or no statutes. A major limitation is the "commercial use" requirement, which means that deepfakes created for non-commercial purposes, such as political propaganda or personal harassment, may not be covered.
Defamation: When a deepfake falsely depicts an individual in a way that harms their reputation, defamation laws (libel for written or visual content, slander for spoken words) can apply. A successful claim requires proving that the content is false, that it caused tangible harm, and that it was published with a degree of fault, such as negligence or malice. However, the legal bar can be high. For example, it can be difficult to prove a deepfake is making a clear "false factual assertion," especially if it's presented as satire or opinion.
Privacy and False Light: The tort of "false light" allows individuals to sue when they are portrayed in a highly offensive or misleading way that causes significant emotional distress. This can be relevant for deepfakes that place individuals in compromising or false situations. However, proving "substantial emotional distress" is a difficult legal standard to meet.
Fraud and Harassment Statutes: In cases where deepfakes are used for criminal purposes, existing laws against wire fraud, computer fraud, or harassment may be applicable. For example, using a deepfaked audio of a CEO to authorize a fraudulent wire transfer would likely fall under wire fraud statutes. Many states have also begun to update their criminal codes to explicitly include deepfakes in laws concerning harassment and exploitation.

While these alternative legal routes provide some avenues for recourse, they highlight the absence of a comprehensive legal tool specifically designed to combat the unique harms of deepfakes. They are often reactive, difficult to prove, and fail to address the underlying issue of the unauthorized creation and distribution of synthetic media itself.

The Legislative Front: A Global Race to Regulate

Recognizing the gaps in existing law, governments around the world have begun to introduce and enact legislation specifically targeting the creation and distribution of malicious deepfakes. These efforts vary widely in their scope and approach, creating an evolving and fragmented global regulatory landscape.

United States: A Federal and State Response

In the U.S., the legislative response has been two-pronged, with a flurry of activity at both the federal and state levels.

Federal Legislation:

Congress has been actively debating several bills aimed at curbing the misuse of deepfakes.

The TAKE IT DOWN Act: Enacted in May 2025, this landmark federal law criminalizes the knowing publication of non-consensual intimate imagery, including AI-generated deepfakes. It mandates that online platforms establish a notice-and-takedown procedure, requiring them to remove flagged content within 48 hours and make efforts to delete duplicates.
The DEEPFAKES Accountability Act: This proposed bill would require creators to digitally watermark or otherwise clearly label deepfake content, ensuring transparency for viewers. It aims to protect national security and provide legal remedies for individuals harmed by unlabeled deepfakes.
The DEFIANCE Act (Disrupt Explicit Forged Images and Nonconsensual Edits Act): This bill would grant victims of non-consensual deepfake pornography the right to sue the perpetrators in federal court for significant damages. Reintroduced in 2025, it aims to provide powerful civil recourse for victims.
The NO FAKES Act: This proposed legislation seeks to make it illegal to produce or distribute an unauthorized AI-generated replica of a person’s voice or likeness, with exceptions for parody and news commentary.
National Defense Authorization Act (NDAA): Updates to the NDAA in 2024 included provisions to address the threat of deepfakes in the context of national security, particularly their potential use in foreign misinformation campaigns.

State-Level Action:

While federal laws are taking shape, numerous states have already passed their own legislation. As of 2024, many states had laws targeting online impersonation and had begun updating them to include deepfakes. These laws often focus on specific harms:

Election Integrity: States like Texas and California have passed laws that prohibit the distribution of deceptive deepfakes within a certain period before an election with the intent to injure a candidate or mislead voters. New York's 2024 law, for instance, requires clear labeling on any AI-altered political material that could be mistaken for the real thing.
Non-Consensual Pornography: New York's Hinchey Act of 2023 made it a crime to create or share sexually explicit deepfakes without consent and gave victims the right to sue. Many other states have expanded their existing laws against non-consensual intimate imagery to explicitly cover AI-generated forgeries.

This state-by-state approach has created a "patchwork" of laws, leading to inconsistencies in protection and enforcement across the country.

European Union: A Focus on Transparency

The European Union has taken a broad, risk-based approach to regulating artificial intelligence, with deepfakes falling under its comprehensive new rules.

The EU AI Act: As a pioneering piece of legislation, the AI Act categorizes AI systems by risk. Under this act, AI systems that generate or manipulate image, audio, or video content (i.e., deepfakes) are subject to specific transparency obligations. Anyone deploying such a system must clearly disclose that the content is artificially generated or manipulated, unless it is part of an obviously creative or satirical work. The goal is to ensure users are aware when they are interacting with synthetic media.
The Digital Services Act (DSA): The DSA complements the AI Act by imposing rules on online platforms for content moderation. It requires platforms to be more transparent about their rules and to have mechanisms in place for users to report illegal content, which would include harmful deepfakes.

United Kingdom: Safety and Security

The UK has integrated deepfake regulation into its broader internet safety legislation.

The Online Safety Act 2023: This act makes it illegal to share sexually explicit deepfake images if it is done with the intent to cause distress, or if the sender is reckless as to whether distress is caused. The law places a significant responsibility on online platforms to remove such illegal content quickly. However, for deepfakes that are not sexually explicit, victims must still often rely on existing laws for defamation or harassment, highlighting potential gaps in the regulatory framework.

China: Proactive and Strict Regulation

China has been one of the most proactive nations in regulating deepfakes and generative AI.

Provisions on the Administration of Deep Synthesis: These rules, which came into effect in 2023, require all AI-generated content and deepfakes to be clearly labeled to avoid public confusion. Furthermore, service providers must verify the real identities of their users and obtain consent before using someone's likeness or voice. This approach demonstrates a strong emphasis on traceability and accountability.

Denmark: A Pioneering Copyright Approach

In a groundbreaking move, Denmark has proposed tackling deepfakes by expanding copyright law itself. The proposed amendment would make it illegal to publicly share a realistic digital recreation of a person's facial features, appearance, or voice without their explicit consent. This "imitation protection" would give ordinary individuals a copyright-like power over their own likeness, allowing them to demand the removal of deepfakes and seek compensation, regardless of the creator's intent. The protection would even extend for 50 years after the person's death. This harm-agnostic, consent-based model is being watched closely as a potential blueprint for other nations.

The Fair Use Battleground: Training AI on Copyrighted Data

A central and fiercely contested issue in the AI era is the legality of using massive datasets of copyrighted material—text, images, music, and videos scraped from the internet—to train generative AI models. AI developers often argue that this practice constitutes "fair use" (or "fair dealing" in some jurisdictions), a legal doctrine that permits the limited use of copyrighted works without a license for transformative purposes like research, criticism, and education.

The argument is that using works for training is transformative because the AI is not reproducing the works but rather learning the underlying patterns, styles, and logic from them to generate something new. However, creators and copyright holders vehemently disagree. They contend that this unauthorized use devalues their work, as the AI models are then used to produce content that directly competes with the very material they were trained on. They argue that if AI developers want to use their work, they should be required to obtain a license and pay for it.

This conflict is now being fought in courtrooms. A wave of lawsuits has been filed by artists, authors, and news organizations against major AI companies. One of the most prominent cases is Getty Images v. Stability AI, in which Getty Images alleges that the AI art generator Stable Diffusion was trained on millions of its copyrighted images without permission, even replicating the Getty Images watermark in some of its outputs. The outcomes of these landmark cases will have profound implications for the future of both the creative industries and AI development, potentially redrawing the boundaries of copyright law for the digital age.

Who Is Liable? Unraveling the Chain of Responsibility

When a harmful or infringing deepfake is created and distributed, determining who is legally responsible is incredibly complex. The "black box" nature of many AI systems, where it's difficult to trace exactly why a specific output was generated, complicates the assignment of blame. Several parties could potentially be held liable:

The AI Developer: The company that built and trained the AI model could be held responsible, particularly if they trained their model on infringing data or designed it in a way that encourages infringing outputs. Some legal scholars argue that AI companies could be held liable for direct copyright infringement.
The User: The individual who enters the prompt that generates the deepfake could be seen as the "author" of the infringing content and therefore be held directly liable for its creation.
The Platform: The social media site, website, or app that hosts and distributes the deepfake could be held liable for its dissemination, especially if they fail to remove it after being notified. Laws like the TAKE IT DOWN Act in the U.S. and the Digital Services Act in the EU are increasingly placing this responsibility on platforms.

Some legal theorists have even proposed a novel solution: treating the AI system itself as an artificial legal person for the purposes of infringement. In this scenario, the AI would be the direct infringer, and the humans involved (the developer and the user) could then be held secondarily liable based on their level of control and involvement. This radical idea highlights the search for new legal paradigms capable of addressing the unique challenges of AI.

The Technological Arms Race and the Path Forward

As deepfake technology becomes more sophisticated, a parallel effort is underway to develop technologies to counter it. Researchers are creating advanced AI-powered detection tools that can spot the subtle imperfections and digital fingerprints left behind during the deepfake creation process. Simultaneously, there is a strong push for implementing digital watermarking and metadata standards, which would embed a permanent, traceable marker in AI-generated content to indicate its origin—a solution that some proposed laws, like the DEEPFAKES Accountability Act, seek to mandate.

However, technology alone is not a panacea. The path forward requires a multifaceted approach that combines robust legislation, technological innovation, international cooperation, and public education. Legislating against deepfakes is a delicate balancing act. Lawmakers must craft regulations that are strong enough to protect individuals from harm, safeguard the integrity of information, and uphold intellectual property rights, all without stifling the innovation and creative potential of artificial intelligence.

The challenge is to create a legal framework that is both specific enough to address the harms of today and flexible enough to adapt to the technology of tomorrow. As nations around the world continue to experiment with different regulatory models—from the broad, risk-based approach of the EU to the strict, consent-based laws in China and the rights-expanding proposals in Denmark—a global consensus may slowly emerge. The journey to effectively regulate the digital frontier is still in its early stages, but it is a critical one in ensuring that the future of AI is one that enhances, rather than erodes, truth, trust, and human dignity.