Researchers from Harvard have been examining the complex legal challenges posed by artificial intelligence (AI)-generated non-consensual intimate imagery, commonly called "deepfakes."
While United States (US) laws against "revenge porn" are now universal, they struggle to address images that are synthetically created rather than real photographs.
The analysis carried out by Harvard explores the current ambiguous legal landscape, raises the central legal question of "who is liable - the toolmaker or the user?", and considers how differing global regulatory philosophies are scrambling to address this rapidly evolving threat to personal dignity and privacy.
Legal Background: A New Frontier in Image-Based Abuse
In 2025, South Carolina became the final state to criminalize revenge porn, formally outlawing the nonconsensual distribution of real sexual images. But the legal system is now facing a much murkier issue around the creation of AI-generated explicit images that are fake, but look disturbingly real.
The US currently lacks comprehensive laws governing the creation or distribution of these deepfakes. Meanwhile, other countries have moved more aggressively.
The UK, for instance, has threatened bans against platforms like X (formerly Twitter) for failing to act, while the EU, China, India, and South Korea have already passed broad regulations targeting AI misuse.
In the US, a Senate bill recently passed, granting deepfake victims the right to sue, but it remains stalled in the House.
As lawmakers debate, a key legal question remains unresolved. Should the liability rest with the person misusing the technology or the developers enabling it?
How Generative AI Amplifies Sexual Harassment
The issue of unwanted sexualization isn’t new, but generative AI has dramatically escalated both its scale and realism. According to Harvard Law School professor Rebecca Tushnet, AI-generated imagery creates a heightened sense of violation.
“The extra realism of AI-generated representations does seem to make a difference to both the injury and intrusion that people feel,” she explains.
The leap in realism transforms what used to be fringe or niche harassment into a widespread, scalable problem. Although major AI platforms restrict sexually explicit outputs, Tushnet notes that it’s relatively easy for someone with modest technical know-how to fine-tune even a "well-behaved" model using explicit datasets. The result is highly convincing, easily produced deepfakes with the power to harm reputations and inflict psychological trauma.
This accessibility creates a legal dilemma for US regulators.
American law is well-equipped to handle deepfakes used to defame, like fabricating a video of a public figure committing a crime, but not as well-suited to address material that’s clearly fake yet still used to degrade or harass.
As Tushnet points out, much of this content doesn’t aim to deceive. Instead, it’s weaponized to humiliate, often targeting women, marginalized groups, or political figures, without pretending to be real.
The First Amendment poses a challenge, but Tushnet argues the deeper issue is cultural. US law tends to frame harm through a consumer lens (economic or reputational), not personal dignity or psychological abuse. Without a concept of “good-faith” regulatory compliance, American courts may push toward making developers responsible for nearly any misuse of their tools.
Global Comparisons: Different Legal Philosophies, Different Outcomes
The international response to AI deepfakes offers a sharp contrast.
In Europe, platform and AI regulation is more proactive. According to Tushnet, their enforcement models allow for discretion. If companies demonstrate strong “guardrails” and good-faith efforts to prevent abuse, they’re often considered in compliance, even if some misuse occurs.
In the US, that kind of flexibility is rare. Instead, laws often rely on rigid standards with little room for intent or effort.
Plaintiffs can file lawsuits if a tool is capable of being misused, regardless of the developer’s preventive measures. This legal rigidity creates a difficult environment for innovation, where toolmakers face legal risk simply for building something powerful.
Currently, US law focuses on post-harm remedies, such as requiring takedowns of intimate deepfakes after victims notify platforms. But it has little to say about proactive responsibility for creating tools that enable abuse.
The debate is now caught between two poles: one side seeks stronger protections against weaponized imagery that disproportionately harms women and minorities; the other warns against chilling effects on political satire, creative expression, or dissent.
The challenge lies in defining what responsibility should look like, especially in a legal system that defaults to “all or nothing” liability.
Conclusion: Bridging the Legal Gap Between Innovation and Protection
The rise of AI-powered deepfakes has exposed serious gaps in legal frameworks worldwide, but nowhere more acutely than in the United States. The core tension remains: how to hold technology developers accountable without stifling innovation or violating free speech.
Tushnet argues that US law, in its current form, is poorly suited to regulate AI tools that cause personal and psychological harm. Without legal frameworks that recognize non-commercial injuries, like loss of dignity or safety, victims remain in legal limbo.
Moving forward, developing clear standards for what constitutes “reasonable guardrails” will be critical.
But perhaps the more urgent task is agreeing on a legal baseline regarding what is actually required of a toolmaker, and when does their responsibility begin?
Until that’s defined, the legal system will continue to lag behind the technology it’s trying to govern.
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.