Inside AI of the ground war against the celebrity Deepfakes

In January, the French woman lost $ 900,000 after they were the victim of fraudsters who were Brad Pitt, who persuaded her to send money for medical treatment. A sophisticated scheme that lasted over a year, presented photos created by artificial intelligence Showing the actor in the hospital bed, it seems to fight kidney cancer. Pitt representatives He later issued a statement By failure that “fraudsters use a strong connection of fans with celebrities.”

These scams can also work on a scale. This month, investigators opened surgery with Tbilisi, Georgia, which used Deepfakes and fabricated news reports to promote cryptocurrency schemes, which feature famous British personalities, including Martin Lewis’s financial expert, DJ Zoe Ball Radio. The operation cheated thousands Victims of $ 35 million.

Welcome to reality around the clock, which has encountered celebrity and glory in these times, without more than Hollywood, loaded with stars.

Although AI is the culprit in many of these cases, weapons to combat this battle also work from AI, emphasizing a new market opportunity for digital beginners who seek to be on advanced lines for studies and talent agencies.

Enter Lottothat calls itself a company “Protection Protection Technology” that has Partnership with the WME Giant Talent Agency and is provided 7 million dollars financing seeds last fall. He positions himself as a solution to a growing problem: the dissemination of AI -generated content, which does not assign, does not allow, faces the face and voices of celebrities.

“Entering this area, you cannot really fight technology without technologies to support what you do”, Chris Jacquin, WME partner and digital strategy manager, said Thewrap last fall.

The size of the problem is stunning. According to DeepMedia, another company that is working on AI detection tools, approximately 500,000 audio and visual deepfakes shared on social networks in 2023. For the largest stars, the impact can be serious. In January 2024, X blocked searches For Taylor Swift after pornographic Deepfakes, the singer has acquired a viral on the platform.

Legislators try to adopt laws that could create consequences for people who make these non -consensual images. US Senators Chris Coner (D-D-Minnesota), Marsha Blackburne (R-Tennesse) and Tom Tillis (R-Nrth Carolina) plan to re-enter the law without forgery during weeks, According to CNN. This legislation joins Take it downwhich is aimed at generating AI-generation of deeply prepared pornography and recently gained support from the first lady of Melania Trump.

“Everyone deserves the right to own and protect their voice and likeness, regardless of whether you are Taylor Swift or someone else,” said senator kuna when not fake was first presented in 2024.

But even if the legislation is adopted, the images will continue to be stored by storing companies such as Loti employed. Luke Arrigoni, the executive director of Loti, describes the ecosystem of unauthorized content, which varies from simply annoying to potentially destructive. The company uses a multi -level approach to deleting requests, recognizing that not all AI media generated requires the same answer.

The most flashy category includes the obvious material that the platform is usually removed immediately after notification. Emphets’ accounts on social media platforms are much more common and complex, which, according to Arrigoni, can do significant economic damage. For one loud customer, Loti can issue 1000 requests for removal per day.

As for fan content, Loti clients apply a more nuanced approach. Some clients distinguish between harmful publications and benign creative content, which leads to different standards for deleting.

“There is a sliding scale,” Arrigoni said. “I would say that half of our customers don’t even care about benign. If it’s not too crazy, they will say that you don’t have to take it off. ”

While Arrigoni refuses to identify specific clients, he acknowledges that Lotie serves people in entertainment, politics and corporate leadership, including military officials, politicians, executives, real estate managers of dead celebrities and numerous musicians, especially in the DACA.

It hints at the registry, which includes “the famous introverted adrenaline drug addict with a tendency to long coats”, “The most trusts Hollywood”, “Philosopher Podcasts, which moonlights as an enthusiast of combat sports” and “hip-hop-icon, whose sign. The company also monitors the accounts of celebrities, such as managers or family members, who often present themselves in difficult fans focused.

100 million scans a day

The company’s approach is technological progress in previous methods in which unauthorized content was determined by a random fan -communicates that warn management teams. These teams then involve lawyers in the deleting-process negotiations, which does not fit the volume and speed of today’s digital environment.

The Lotie System scans about 100 million images and videos daily using 40 AI models in six services to detect unauthorized content. Their methods include facial recognition, voice recognition and representation, and the company requires 95% success in removing materials that are violated within 17 hours.

Instead of using traditional fingerprints that exhibit copyright violations but become ineffective when the faces change or change backgrounds, the lotte technology is actually more accurate, since the content of AI is increasing.

“The better the technology becomes, the more realistic it looks, the easier our things should work to find it,” he said.

For a standard monthly fee of $ 2500, customers receive unlimited removal requests – a pricing model that deliberately avoids the fine of those who face disproportionate targeting, especially women.

“These people are victimized,” Arrigoni said. “We do not try to be combat profit.”

Arrigoni recognizes the paradox inherent in the Loti business model: the use of AI to combat the harmful effects of the same technology.

“I sell gold pots and gold elections,” he said, referring to gold rushs when there is always a sub-industry that makes money in elections and shovels. “Is this bizarre irony, right?”

The company faces the fact that irony is heading, explaining its methodology to the legal teams of clients. The process requires confidential permits – scanning facial features and processing biometric data with AI models – which can cause anxiety. Arrigoni said Loti appeals to them, emphasizing strict restrictions on data use, and offering a contract that the company would never create generative models of its customers.

However, he argues that this approach is a pragmatic and ethical use of tools that could otherwise be laid out only by human efforts.

Russia: outside

Despite the fact that Loti technology can effectively counteract AI threats in most of the Internet, it cannot overcome geopolitical realities. Despite the complex tools of the company and wide coverage, Arrigoni recognizes that international borders are still representing barriers.

“We are quite understandable with customers – Russia does not go beyond,” he said.

Russian sites tend to ignore the removal requests. Removal of lottoes in several layers of online infrastructure – orientation to not only web sites, but also domain registrators, content networks and even processors of payment – but Russian platforms have developed independent alternatives to these services, which provides such approaches ineffective. However, Loti is capable of involving social media posts related to the taken sites, which reduces their visibility and accessibility to the average user.

For public figures faced with the flow of unauthorized digital ideas – from explicit fictional accounts aimed at their supporters – such technologies offer some more and more complex digital landscape. Many clients report that in addition to technical success, the service provides psychological relief from constant monitoring of their digital presence.

“We get a consistent response that a set of problems is getting smaller,” he said. “The problem is solved for people when they join us. The mental space dedicated to this disappeared. “

Post Inside AI of the ground war against the celebrity Deepfakes appeared first on Terop.