Regulating Generative AI for Images and Video

This page is a translation of the original Portuguese text, provided for convenience. It deals with Brazilian law. If there is any discrepancy, the Portuguese version prevails. Read the original in Portuguese

Generative artificial intelligence for images and video is advancing at an unprecedented pace, and Brazil faces the urgent challenge of building a regulatory framework that protects fundamental rights without stifling technological innovation.

The Current Landscape of Generative AI for Visual Content

We are living through a profound shift in how images and videos are created, distributed, and consumed. Generative AI tools allow anyone to produce realistic visual content from simple text prompts, including synthetic photographs, complex illustrations, videos featuring human faces, and even re-creations of historical settings. Only a few years ago, this capability seemed confined to science fiction. Today it is available on commercial platforms used by millions of people around the world.

This democratization of visual creation clearly brings benefits to fields such as advertising, education, entertainment, and corporate communications. Still, the risks of indiscriminate use of these technologies are just as significant. Deepfakes (fabricated videos that look real), non-consensual intimate images, the infringement of artists’ and photographers’ copyrights, and the mass production of visual disinformation all pose concrete threats to human dignity, legal certainty, and the functioning of democracy.

In Brazil, the debate over AI regulation has gained momentum as bills moving through the National Congress seek to establish general principles for the development and use of AI systems. Although these bills address artificial intelligence broadly, regulation aimed specifically at generative AI for visual content still lacks detailed rules. That creates uncertainty for developers, for users, and for people who may be affected by these technologies.

Fundamental Rights at Risk: Image, Privacy, and Intellectual Property

Article 5 of the 1988 Federal Constitution guarantees the inviolability of people’s intimacy, private life, honor, and image. The Civil Code, in turn, reinforces the protection of the right to one’s image as a personality right. When a generative AI tool creates a video or photograph using a person’s face without authorization, there is a potential violation of these fundamental rights, regardless of whether the content was produced by an algorithm rather than a camera.

The issue becomes even more complex when we consider the General Data Protection Law (LGPD). Facial biometric data used to train generative AI models is classified as sensitive personal data, which requires the data subject’s specific and prominent consent. In practice, many generative AI systems were trained on billions of images collected from the internet without any form of consent. This raises serious questions about the lawfulness of those datasets under Brazilian law.

The situation is equally delicate in the field of intellectual property. The Copyright Law protects original works resulting from human creation. When an AI system generates an image that reproduces a specific creator’s artistic style, or uses fragments of protected works in its synthesis process, copyright questions arise that current law does not answer unambiguously. Artists and photographers have voiced legitimate concerns about the unauthorized use of their work to train generative models, arguing that the practice amounts to misappropriation of their creative effort.

Deepfakes and Human Dignity

Among the most troubling applications of generative AI for video are so-called deepfakes: manipulated audiovisual content that can place anyone in fictional situations with a high degree of realism. The use of this technology to create non-consensual intimate content (known as deepfake pornography) already affects thousands of victims, predominantly women, and is a serious form of digital violence. Brazil’s Criminal Code has been updated to criminalize conduct involving the manipulation of intimate images, but the speed at which these tools evolve demands constant legislative vigilance.

Regulating generative visual AI is not just a technology question. It is an urgent necessity for protecting fundamental rights such as image, privacy, and human dignity in the digital environment.

Beyond the individual sphere, politically motivated deepfakes pose a direct threat to the democratic process. Fake videos attributed to candidates or public officials can distort electoral debate and undermine trust in institutions. The Electoral Courts have sought ways to curb these practices, but how easily such content can be produced and spread creates significant operational challenges.

Regulatory Proposals in Brazil and Around the World

Internationally, the European Union stands out with its AI Act (the European Artificial Intelligence Regulation), which sets specific obligations for generative AI systems, including transparency about artificially generated content and mandatory labeling of deepfakes. This regulatory model takes a risk-based approach, sorting AI systems into categories according to their potential to cause harm, with requirements proportionate to each level.

In the United States, regulation has advanced in a fragmented way, through state-level and sector-specific initiatives addressing particular issues such as election deepfakes and the protection of minors. China, by contrast, has adopted rules that require AI-generated content to be identified and that assign responsibilities to providers of generative services, taking a more interventionist stance.

In Brazil, the bills under discussion in the National Congress seek to establish a general regulatory framework for artificial intelligence, built on principles such as transparency, explainability, safety, and non-discrimination. Some of these bills contain specific provisions on synthetic content, such as mandatory labeling of AI-generated images and videos and a ban on deepfakes in electoral contexts. Detailed sector-specific regulation, however, still needs further legislative work. That includes civil liability for damage caused by generated content, technical traceability requirements, and the obligations of platform providers.

The Role of the ANPD and Other Regulators

The ANPD (Brazil’s National Data Protection Authority) plays a central role in overseeing how generative AI systems use personal data. We believe the ANPD has the authority to require developers and operators of generative tools to demonstrate the legal basis for processing the biometric data and personal images used to train their models. In addition, other bodies, such as the Public Prosecutor’s Office, the Procons (consumer protection agencies), and CADE (Brazil’s antitrust authority), can act within their respective areas of authority to curb abuses involving these technologies.

Cooperation among different regulators will be essential for effective governance of generative visual AI. The risks involved are multifaceted, spanning data protection, consumer rights, intellectual property, election law, and criminal law. They call for a coordinated approach that avoids both regulatory gaps and unnecessary overlaps in authority.

Civil Liability and the Challenges Facing Victims of AI-Generated Content

One of the most sensitive points in regulating generative AI for images and video is assigning civil liability when synthetic content harms third parties. Brazilian law does offer a basis for holding someone liable, notably through Articles 186 and 927 of the Civil Code (liability for wrongful acts) and the Consumer Protection Code (strict liability of the supplier for defects in a product or service). Even so, the chain of actors involved in generating and spreading AI content makes identifying the responsible party a complex task. That chain includes the model’s developers, the platform providers, the users who write the prompts, and the platforms that host the final content.

The victim of a deepfake or an improperly generated image faces significant practical obstacles. Identifying who created the content can be hampered by platform anonymity, the absence of usage records, or the fact that service providers are based in other countries. Producing technical evidence that the content is artificial requires specialized expert analysis, which can be prohibitively expensive for many people. And viral content spreads so quickly online that a court order obtained after lengthy proceedings is often too late to be effective.

We believe Brazilian law needs to evolve in three ways. It should establish fast mechanisms for the removal of AI-generated content that violates the rights of others. It should define the chain of responsibility among the various actors involved. And it should create procedural tools that effectively protect victims. Requiring providers of generative tools to maintain traceability and usage records would be an important step toward identifying those responsible and gathering evidence in any eventual lawsuit.

Paths Toward Balanced Regulation

We recognize that regulating generative AI for visual content must strike a delicate balance between protecting fundamental rights and preserving freedom of expression and technological innovation. An overly restrictive approach could stifle the development of legitimate, beneficial applications, while a lack of adequate regulation exposes individuals and society to unacceptable risks.

A few principles should guide this regulatory process. First, transparency: users and affected individuals should have the right to know when visual content was generated or manipulated by artificial intelligence, through clear labeling and verifiable metadata. Second, proportionality: regulatory requirements should match the actual risk of each application, distinguishing artistic and educational uses from potentially harmful ones. Third, effective accountability: the chain of responsibility should be clearly defined, so that victims have access to swift protection and redress.

It is also important to invest in digital literacy, so that people can recognize synthetic content and think critically about visual material of uncertain origin. Digital education, together with technological tools for detecting AI-generated content and a solid regulatory framework, forms an essential tripod for meeting the challenges this technological revolution poses for the law and for Brazilian society.

This topic is closely related to artificial intelligence and regulation, which we cover in a dedicated article.

This topic is closely related to AI regulation in the financial sector, which we cover in a dedicated article.

As informações deste artigo são de caráter informativo e não substituem consulta jurídica individualizada. Cada caso possui particularidades que exigem análise específica.