Nonconsensual deepfake porn is no longer a fringe abuse of synthetic media. The recent spread of sexually explicit AI-generated images of Taylor Swift on X, formerly known as Twitter, showed how fast this material can move and how difficult it can be for platforms to contain once it is public.
The problem is not new, but generative AI has made it easier to create convincing images and videos. Henry Ajder, an AI expert who specializes in generative AI and synthetic media, says nonconsensual deepfakes affect the largest number of people among harms linked to generative AI, with women making up the vast majority of those targeted.
Why the Taylor Swift case matters
Millions of people viewed nonconsensual deepfake porn of Taylor Swift on X. The platform later took the drastic step of blocking all searches for Taylor Swift as it tried to control the spread.
That response points to a larger challenge. Platforms already review uploaded posts and remove material that violates their rules, but moderation is inconsistent and can miss harmful content. Deepfake porn makes that harder because platforms must also determine whether an image is authentic or AI-generated.
The stakes are personal and practical. Victims need faster removal, clearer accountability and better ways to prevent their images from being weaponized. Platforms need tools that can identify synthetic content at scale. Lawmakers are under pressure to give victims more recourse and create consequences for people who create or share this material.
Watermarks could help platforms find AI images faster
One technical response is watermarking. A watermark can place an invisible signal inside an image so that computers can detect whether it was generated by AI. Google has developed a system called SynthID, which uses neural networks to modify pixels in images in a way that is invisible to people.
The promise is straightforward: if AI-generated images can be flagged reliably, platforms may be able to spot nonconsensual deepfakes faster and remove them sooner. Sasha Luccioni, a researcher at the AI startup Hugging Face who has studied bias in AI systems, says applying watermarks to all images by default would also make it harder for attackers to create nonconsensual deepfakes in the first place.
But watermarking is not a complete answer. These systems remain experimental and are not widely used. A determined attacker may still find ways to interfere with them. Companies are also not applying watermarks across all images by default. For example, users of Google’s Imagen AI image generator can choose whether their AI-generated images have the watermark.
That weakens the value of watermarking as a universal defense. A tool that only covers some systems and some outputs can help moderation, but it cannot stop deepfake porn on its own.
Protective shields can make images harder to exploit
Another line of defense focuses on the images people post online. Today, photos uploaded to the internet can be used by others to create deepfakes. As image-making AI systems become more sophisticated, proving that manipulated content is fake is also becoming more difficult.
Several defensive tools try to change that by altering images in ways people cannot see but AI systems can. PhotoGuard, developed by researchers at MIT, changes pixels in photos so that if an AI app like the image generator Stable Diffusion is used to manipulate the image, the output looks unrealistic.
Fawkes, developed by researchers at the University of Chicago, uses hidden signals that make faces harder for facial recognition software to recognize. Nightshade, also developed by researchers at the University of Chicago, applies an invisible layer of “poison” to images. It was built to protect artists from having copyrighted images scraped by tech companies without consent, but in theory it could be used on any image an owner does not want scraped by AI systems.
These tools could be especially useful if dating apps and social media companies apply them by default, Ajder says. Luccioni puts the case more directly:
“We should all be using Nightshade for every image we post on the internet,” says Luccioni.
The limits are just as important. These defenses work against the latest generation of AI models, but future systems may be able to get around them. They also do not protect photos that are already online. And they are harder to apply to celebrities, because public figures do not control every photo of themselves that others upload.
Rumman Chowdhury, who runs the ethical AI consulting and auditing company Parity Consulting, describes the dynamic as an ongoing contest:
“It’s going to be this giant game of cat and mouse,” says Rumman Chowdhury, who runs the ethical AI consulting and auditing company Parity Consulting.
Regulation may decide whether deterrence is real
Technical tools can reduce harm, but they do not settle the accountability question. Luccioni argues that lasting change will require stricter regulation.
The Taylor Swift incident has renewed attention on efforts to curb deepfake porn. The White House said the incident was “alarming” and urged Congress to take legislative action. In the US, the current approach has been piecemeal and state by state. California and Virginia have banned the creation of pornographic deepfakes made without consent. New York and Virginia also ban distribution of this content.
There may also be movement at the federal level. A new bipartisan bill that would make sharing fake nude images a federal crime was recently reintroduced in the US Congress. A deepfake porn scandal at a New Jersey high school has also pushed lawmakers to respond with a bill called the Preventing Deepfakes of Intimate Images Act.
Other regions are taking different approaches. The UK’s Online Safety Act, passed last year, outlaws the sharing of deepfake porn material, though not its creation. Perpetrators could face up to six months of jail time. In the European Union, the AI Act requires deepfake creators to clearly disclose that material was created by AI, while the Digital Services Act will require tech companies to remove harmful content much more quickly.
China’s deepfake law, which entered into force in 2023, goes further. In China, deepfake creators need to prevent illegal or harmful use of their services, ask for consent before turning people’s images into deepfakes, authenticate identities and label AI-generated content.
No single fix is enough
The clearest path is a combination of defenses. Watermarks can help platforms identify AI-generated material. Protective shields can make personal images harder to exploit. Regulation can give victims recourse, hold creators accountable and send a clearer message that nonconsensual deepfake pornography is a form of sexual abuse.
Ajder says public awareness campaigns and laws that make it clear people who create this content are sex offenders could have a real impact. He warns, however, that enforcement will be difficult. Victims may struggle to identify who created the content and build a case, and the person responsible may be in a different jurisdiction.
That means the response must be practical rather than symbolic. Platforms, AI companies, app makers and lawmakers each control a different part of the problem. The Taylor Swift case showed the scale of the harm. The next test is whether the tools and rules now being discussed can make deepfake porn harder to create, faster to remove and riskier to spread.