Microsoft Launches Tool to Combat Deepfake Porn in Bing Search Results

Microsoft gives deepfake porn victims a tool to scrub images from Bing search | TechCrunch

The advancement of generative AI tools has created a new problem for the internet: the proliferation of synthetic nude images resembling real people. On Thursday, Microsoft took a major step to give revenge porn victims a tool to stop its Bing search engine from returning these images. Microsoft announced a partnership with StopNCII, an organization that allows victims of revenge porn to create a digital fingerprint of these explicit images, real or not, on their device. StopNCII's partners then use that digital fingerprint, or "hash" as it's technically known, to scrub the image from their platforms. Microsoft's Bing joins Facebook, Instagram, Threads, TikTok, Snapchat, Reddit, PornHub, and OnlyFans in partnering with StopNCII, and using its digital fingerprints to stop the spread of revenge porn. In a blog post, Microsoft says it already took action on 268,000 explicit images being returned through Bing's image search in a pilot through the end of August with StopNCII's database. Previously, Microsoft offered a direct reporting tool, but the company says that's proven to be not enough. "We have heard concerns from victims, experts, and other stakeholders that user reporting alone may not scale effectively for impact or adequately address the risk that imagery can be accessed via search," said Microsoft in its blog post on Thursday. You can imagine how much worse that problem would be an a significantly more popular search engine: Google. Google Search offers its own tools to report and remove explicit images from its search results, but has faced criticism from former employees and victims for not partnering with StopNCII, according to a Wired investigation. Since 2020, Google users in South Korea have reported 170,000 search and YouTube links for unwanted sexual content, Wired reported. The AI deepfake nude problem is already widespread. StopNCII's tools only work for people over 18, but "undressing" sites are already creating problems for high schoolers around the country. Unfortunately, the United States doesn't have an AI deepfake porn law to hold anyone accountable, so the country is relying on a patchwork approach of state and local laws to address the issue. San Francisco prosecutors announced a lawsuit in August to take down 16 of the most "undressing" sites. According to a tracker for deepfake porn laws created by Wired, 23 American states have passed laws to address nonconsensual deepfakes, while nine have struck proposals down.

Microsoft removes revenge porn from Bing search using new tool

Microsoft announced today that it has partnered with StopNCII to proactively remove harmful intimate images and videos from Bing using digital hashes people create from their sensitive media. StopNCII is a project operated by the Revenge Porn Helpline that allows people to create digital hashes of their intimate pictures and videos without uploading the media from their phone. StopNCII then adds these hashes to a database used to find the same or similar images online, which are then removed by their partners, including Facebook, TikTok, Reddit, Pornhub, Instagram, OnlyFans, and Snapchat. In March, Microsoft shared its PhotoDNA technology with StopNCII, allowing enhanced creation of digital hashes without a person's images or videos leaving their device. "Much like the hashing technology that StopNCII.org already uses (PDQ), PhotoDNA is an additional process that enables identified harmful images to be hashed into a digital fingerprint, which then can be shared with industry platforms to identify and remove any non-consensual intimate image abuse material," explains the March announcement. Microsoft announced today that they have been piloting the use of StopNCII's database of hashes to remove intimate images within the Bing search index. Using this database, Microsoft says they have taken action on 268,899 images through the end of August. The rise of artificial intelligence has also led to an increased generation of deep fake nude images from non-intimate photos shared online. While these images are fake, they can be as equally distressing to those who are being exploited. A 2019 report by DeepTrace, now Sensity, showed that 96% of deepfake videos on the internet are pornographic in nature and almost all featured non-consensual use of a woman's likeness. Many of these images are uploaded as "revenge porn," for extortion, or to generate revenue by unscrupulous sites. Unfortunately, AI-generated images make it harder to match against PhotoDNA hashes. In these cases, those impacted should manually report the images to Microsoft, Google, and other online media companies. Microsoft says that impacted people can use their Report a Concern page to request that real or synthetic images be taken out of the Bing search index.