A few weeks ago, a Google search for “deepfake nudes jennifer aniston” turned up at least seven high-level results that purported to feature explicit, AI-generated photos of the actress. Now they’re gone.
Emma Higham, Google’s product manager, says modern changes to the way the company ranks results, which were introduced this year, have already reduced exposure to fraudulent sexual images by more than 70 percent in searches for content about a specific person. Where there might once have been problematic results, Google’s algorithms are designed to promote news articles and other non-sexual content. The Aniston search now returns articles like “How Taylor Swift’s Deepfake AI Porn Poses a Threat“and other links like Ohio Attorney General’s Warning on “celebrity endorsement deepfake scams” targeting consumers.
“These changes will allow people to read about the impact deepfakes have on society, rather than seeing pages of real, unwanted, fake images,” Higham wrote in a company blog post Wednesday.
The ranking change follows an investigation this month by WIRED, which revealed that Google executives in recent years have rejected numerous ideas put forward by employees and outside experts to combat the growing problem of intimate images of people being shared online without their consent.
While Google has made it easier to request removal of unwanted pornographic content, victims and advocates have called for more proactive steps. But the company has tried to avoid becoming too much of a regulator of the internet or making it harder to access legal pornography. At the time, a Google spokesperson said in response that multiple teams are difficult at work strengthening protections against what it calls nonconsensual pornographic images (NCEI).
The increasing availability of AI image generators, some with few restrictions on their employ, has led to the rise of NCEI, according to victim advocates. These tools have made it basic for almost anyone to create fraudulent, explicit images of anyone, whether they’re a high school classmate or a mega-celebrity.
In March, a WIRED analysis found that Google received more than 13,000 requests to remove links to a dozen of the most popular websites hosting explicit deepfakes. Google removed the results in about 82 percent of cases.
As part of Google’s modern actions, Higham says the company will begin implementing three measures designed to reduce the detection of real but unwanted explicit images to those that are synthetic and unwanted. After Google grants a request to remove a sexual deepfake, it will try to exclude duplicates from results. It will also filter explicit images from results for queries similar to those listed in the removal request. Finally, websites that are the subject of a “high number” of successful removal requests will be demoted in search results.
“These actions are intended to provide people with greater peace of mind, especially if they are concerned about similar content appearing about them in the future,” Higham wrote.
Google has acknowledged that the measures aren’t working perfectly, and former employees and advocates for victims have said they could go much further. The search engine explicitly warns people in the US who search for nude images of children that such content is illegal. The effectiveness of the warning is unclear, but it’s a potential deterrent advocates support. But despite the rules against sharing NCEI, similar warnings don’t appear for searches for adult deepfakes. A Google spokesperson confirmed that won’t change.