Photo from Polls in USA
As current artificial intelligence makes groundbreaking progress day by day, it is reshaping the information landscape in ways that threaten the very foundations of democratic integrity. Deepfakes are AI content that allow users to create realistic-looking video, audio, and images of people or events that never occurred. The ability to create deepfake videos requires very little technical expertise and has opened up a world of opportunities to produce highly believable representations of political candidates, public officials, and elections in particular. So as these tools become more accessible, concerns about their effects on democracy have intensified.
Despite risks associated with deepfakes that could impact elections at the federal level, federal regulations pertaining to the use of deepfakes in political campaigns are extremely scarce. Although the Take it Down Act is a federal bill passed in 2025 to reduce deepfake circulation, it targets all forms of nonconsensual media and lacks a focus on deepfake disinformation specifically. Most of the legislation regulating the use of deepfakes emerged at the state level, but deepfakes do not adhere to state boundaries. Deepfakes generated in one jurisdiction can instantly circulate nationwide. Differing state standards consequently create regulatory gaps and compliance challenges for platforms, developers, and users. Fragmented regulation also weakens deterrence. A cohesive federal regulatory structure would establish uniform expectations, reduce jurisdictional inconsistencies, and provide clearer guidance to relevant actors.
Concerns about election disinformation are not new. False statements, misleading imagery, and persuasive distortions have long been features of political communication. Since many of these sources of disinformation are protected under free speech, it leads to debate on whether restricting deepfakes during an election cycle would violate the First Amendment’s free speech clause. Unlike traditional forms of disinformation, however, deepfakes exploit the tendency of viewers to treat audiovisual media as reliable representations of reality. When individuals see a video or hear an audio recording, they often assume authenticity. Deepfakes manipulate this assumption by presenting fabricated events with persuasive credibility. The resulting harm extends beyond reputational damage. Deepfakes can influence voter perceptions during critical decision-making periods. For instance, in the 2024 presidential election black voter support for the Biden Administration dropped by 21% due to deepfake circulation of Donald Trump acting friendly with black communities. In such circumstances, uninformed voters’ judgments are based on false information which furthers concerns of the risks deepfakes pose to electoral and democratic integrity.
Our country’s long held view is that free speech is integral, with political speech being granted the greatest amount of protection when judicially evaluated. Deepfake is an expressive medium, making some view regulation of deepfake speech an infringement upon free speech. However, the First Amendment does not protect all types of expression from government regulation. Fraudulent or defamatory forms of expression have historically been regulated prior to the emergence of deepfake technology. Deepfake technology is considered an expressive medium, but because it is inherently deceptive, determining how the First Amendment protections should apply to deepfake language has become increasingly challenging.
The controversy surrounding Grok, the AI chatbot developed by xAI, illustrates this tension. Some argue that restricting AI-generated content amounts to unconstitutional interference with free speech, while others contend that algorithmically amplified falsehoods fall outside the core protections the First Amendment was designed to safeguard. This divide highlights a growing legal and philosophical dilemma.
A common misconception is that regulating deepfakes requires banning them outright. In reality, constitutionally grounded regulation emphasizing transparency is very effective by helping the public distinguish if something is AI or real. The primary concern with deepfake media is not its existence. It is that audiences often cannot distinguish it from authentic content. Developers of generative AI systems are uniquely positioned to embed warning labels within AI-generated media. Watermarking technologies already demonstrate technical feasibility. Systems such as Google DeepMind’s SynthID and Meta’s Stable Signature illustrate how persistent identifiers can be incorporated into synthetic outputs. Industry reporting has noted that watermarking capabilities exist but have not been uniformly deployed. 30% of users claimed they would stop using ChatGPT if it started implementing watermarks, which is among the reasons why many AI companies do not fully employ labelling technology. Mandating watermarking requirements primarily regulates product design rather than expressive content. This distinction is legally significant because courts have historically applied different standards to design regulations than to direct speech restrictions.
Another way to distinguish deepfakes is through the platforms that circulate them. Social media platforms represent the primary channels for deepfake dissemination. Regulation could also look like platforms detecting watermark labels and providing a warning caption for viewers. In this case, disclosure obligations provide audiences with contextual information about the authenticity of the content being viewed without outright restricting deepfake content. In other words, critiques of watermarking limitations suggest that transparency tools should complement broader safeguards rather than operate in isolation.
Deepfake regulation is frequently implicated as a struggle between freedom of speech and government intervention. However, this conflict can be avoided if it was instead viewed as a way of informing the public rather than censoring deepfakes, as long as it is clear whether content is authentic or AI. The First Amendment protects people’s right to engage in the democratic process but it does not create a free and unrestricted opportunity to create fictitious forms that are intended to deceive individuals who are voting. Federal regulations, AI watermarking, and social media AI warnings can work together to regulate the production and distribution of deepfakes. This integrated force would promote effective discourse among citizens through means of adequate information.
The question at hand is not whether regulation threatens free speech. It is whether democracy in this country can function without reliable systems for distinguishing authentic content from AI fabrication.
• • •
Christina Li is a freshman, studying Business Administration. She is also involved with Women in Business, ScottyLabs, Carnegie AI Safety Initiative, and other organizations.
Leave a Reply