Last week, sexually explicit images of Taylor Swift, one in every of the world’s biggest pop stars, went viral online. Tens of millions of individuals viewed nonconsensual deepfake porn of Swift on the social media platform X, formerly often called Twitter. X has since taken the drastic step of blocking all searches for Taylor Swift to attempt to get the issue under control.
This isn’t a brand new phenomenon: deepfakes have been around for years. Nevertheless, the rise of generative AI has made it easier than ever to create deepfake pornography and sexually harass people using AI-generated images and videos.
Of every type of harm related to generative AI, nonconsensual deepfakes affect the most important number of individuals, with women making up the overwhelming majority of those targeted, says Henry Ajder, an AI expert who makes a speciality of generative AI and artificial media.
Thankfully, there’s some hope. Latest tools and laws could make it harder for attackers to weaponize people’s photos, and so they could help us hold perpetrators accountable.
Listed below are 3 ways we will combat nonconsensual deepfake porn.
WATERMARKS
Social media platforms sift through the posts which are uploaded onto their sites and take down content that goes against their policies. But this process is patchy at best and misses plenty of harmful content, because the Swift videos on X show. It’s also hard to differentiate between authentic and AI-generated content.
One technical solution could possibly be watermarks. Watermarks hide an invisible signal in images that helps computers discover in the event that they are AI generated. For instance, Google has developed a system called SynthID, which uses neural networks to switch pixels in images and adds a watermark that’s invisible to the human eye. That mark is designed to be detected even when the image is edited or screenshotted. In theory, these tools could help firms improve their content moderation and make them faster to identify fake content, including nonconsensual deepfakes.
Pros: Watermarks could possibly be a great tool that makes it easier and quicker to discover AI-generated content and discover toxic posts that must be taken down. Including watermarks in all images by default would also make it harder for attackers to create nonconsensual deepfakes to start with, says Sasha Luccioni, a researcher on the AI startup Hugging Face who has studied bias in AI systems.
Cons: These systems are still experimental and never widely used. And a determined attacker can still tamper with them. Corporations are also not applying the technology to all images across the board. Users of Google’s Imagen AI image generator can select whether or not they want their AI-generated images to have the watermark, for instance. All these aspects limit their usefulness in fighting deepfake porn.
PROTECTIVE SHIELDS
In the meanwhile, all the pictures we post online are free game for anyone to make use of to create a deepfake. And since the most recent image-making AI systems are so sophisticated, it’s growing harder to prove that AI-generated content is fake.
But a slew of recent defensive tools allow people to guard their images from AI-powered exploitation by making them look warped or distorted in AI systems.
One such tool, called PhotoGuard, was developed by researchers at MIT. It really works like a protective shield by altering the pixels in photos in ways which are invisible to the human eye. When someone uses an AI app just like the image generator Stable Diffusion to govern a picture that has been treated with PhotoGuard, the result will look unrealistic. Fawkes, the same tool developed by researchers on the University of Chicago, cloaks images with hidden signals that make it harder for facial recognition software to acknowledge faces.
One other recent tool, called Nightshade, could help people fight back against getting used in AI systems. The tool, developed by researchers on the University of Chicago, applies an invisible layer of “poison” to photographs. The tool was developed to guard artists from having their copyrighted images scraped by tech firms without their consent. Nevertheless, in theory, it could possibly be used on any image its owner doesn’t need to find yourself being scraped by AI systems. When tech firms grab training material online without consent, these poisoned images will break the AI model. Images of cats could turn out to be dogs, and pictures of Taylor Swift could also turn out to be dogs.
Pros: These tools make it harder for attackers to make use of our images to create harmful content. They show some promise in providing private individuals with protection against AI image abuse, especially if dating apps and social media firms apply them by default, says Ajder.
“We must always all be using Nightshade for each image we post on the web,” says Luccioni.
Cons: These defensive shields work on the most recent generation of AI models. But there isn’t a guarantee future versions won’t give you the option to override these protective mechanisms. Additionally they don’t work on images which are already online, and so they are harder to use to photographs of celebrities, as famous people don’t control which photos of them are uploaded online.
“It’s going to be this giant game of cat and mouse,” says Rumman Chowdhury, who runs the moral AI consulting and auditing company Parity Consulting.
REGULATION
Technical fixes go only up to now. The one thing that can result in lasting change is stricter regulation, says Luccioni.
Taylor Swift’s viral deepfakes have put recent momentum behind efforts to clamp down on deepfake porn. The White House said the incident was “alarming” and urged Congress to take legislative motion. Up to now, the US has had a piecemeal, state-by-state approach to regulating the technology. For instance, California and Virginia have banned the creation of pornographic deepfakes made without consent. Latest York and Virginia also ban the distribution of this kind of content.
Nevertheless, we could finally see motion on a federal level. A brand new bipartisan bill that will make sharing fake nude images a federal crime was recently reintroduced within the US Congress. A deepfake porn scandal at a Latest Jersey highschool has also pushed lawmakers to reply with a bill called the Stopping Deepfakes of Intimate Images Act. The eye Swift’s case has delivered to the issue might drum up more bipartisan support.
Lawmakers all over the world are also pushing stricter laws for the technology. The UK’s Online Safety Act, passed last 12 months, outlaws the sharing of deepfake porn material, but not its creation. Perpetrators could withstand six months of jail time.
Within the European Union, a bunch of recent bills tackle the issue from different angles. The sweeping AI Act requires deepfake creators to obviously disclose that the fabric was created by AI, and the Digital Services Act would require tech firms to remove harmful content far more quickly.
China’s deepfake law, which entered into force in 2023, goes the furthest. In China, deepfake creators have to take steps to forestall the usage of their services for illegal or harmful purposes, ask for consent from users before making their images into deepfakes, authenticate people’s identities, and label AI-generated content.
Pros: Regulation will offer victims recourse, hold creators of nonconsensual deepfake pornography accountable, and create a strong deterrent. It also sends a transparent message that creating nonconsensual deepfakes isn’t acceptable. Laws and public awareness campaigns making it clear that folks who create this kind of deepfake porn are sex offenders could have an actual impact, says Ajder. “That will change the marginally blasé attitude that some people have toward this type of content as not harmful or not an actual type of sexual abuse,” he says.
Cons: It’s going to be difficult to implement these forms of laws, says Ajder. With current techniques, it’ll be hard for victims to discover who has assaulted them and construct a case against that person. The person creating the deepfakes may also be in a special jurisdiction, which makes prosecution harder.