New Research Finds 96% of Deepfakes Are Pornographic

Hint: it’s NSFW.

Deeptrace Labs, which builds synthetic media detection tools, recently found that 96% of online deepfake videos are pornographic. Across an ecosystem of websites, anonymous users are creating and spreading these videos, which are mainly shots of female celebrities' faces stitched onto the bodies of adult entertainment stars.

From streaming to VR, the porn industry is often an early adopter of technology. But with deepfakes, this is happening illicitly and without consent. It's exclusively targeting and harming women, Deeptrace wrote.

States are taking this seriously

Last week, California passed a pair of laws taking on deepfakes in pornography and politics.

• The first gives Californians the right to sue someone who creates pornographic deepfakes of their likeness without consent. Virginia also recently expanded its nonconsensual porn ban to include deepfakes.
• The other outlaws deepfakes targeting political candidates within 60 days of an election.

Upshot: With its large economy and progressive state legislature, California has become the U.S.' de facto tech regulator. That means its deepfake laws could reverberate beyond state borders.

Zoom out

Deepfakes burst onto the scene three years ago. But the underlying technology is already being commoditized as AI-generated videos quickly spread across the web. Deeptrace found deepfakes doubled between Dec. 2018 and July 2019, from 7,964 to 14,678.

Experts are worried deepfakes could be a superweapon for propaganda peddlers and misinformation meme makers. Worse, a firehose of deepfakes could sow distrust in institutions, manipulate political processes, and become a credit risk for companies. So Big Tech, Congress, and even the Pentagon are thinking about ways to stop deepfakes from overrunning social media platforms and the internet.

Key questions: Will new regulation blunt the spread of pornographic deepfakes? How will companies and countries combat deepfakes of their CEOs and politicians? Could the response to Question No. 1 help answer No. 2?