Facebook Is Failing To Detect AI Nudes

Plus: U.S. Accuses Chinese AI Companies Of Copying America’s Frontier Models

Forbes

In May this year, a man who was aggrieved at the Tazewell County Sheriff’s Office in Virginia for firing his girlfriend repeatedly posted AI-generated nude images of the agency’s chief and a deputy on Facebook, according to a search warrant reviewed by Forbes.


The original image was taken at an office Christmas party and altered with AI to appear to show the two victims naked, per the warrant. After a Tazewell officer saw the image on his newsfeed, the agency alerted Facebook to the issue and the image was removed. The man continued to post the same image on his public profile and on a story, only for them to be swiftly removed. He also posted the same image, but included a surprised face emoji covering the AI-generated private parts of the victims, and that image, according to the warrant, was not immediately removed, despite the cops reporting it to Facebook.


In another case from earlier this year, a Facebook user discovered someone had turned images of her as a child into child sexual abuse material (CSAM), and shared them via a public Facebook profile. Some of the CSAM was generated from an image of the victim at Disney World, which her father posted on Facebook in 2014 when she was 11 years old. Per the warrant, investigators in Utah obtained data from Meta that showed the alleged perpetrator had made 56 AI-modified pornographic images of the victim from when she was a minor, though it’s unclear how many he posted online.


The warrants highlight the rise of so-called AI “nudifiers” as a harassment tool. They also show that, despite a general ban on nudity across Facebook and attempts to counter AI nudifiers, Meta’s own artificial intelligence isn’t always able to detect AI-generated nudes when they’re posted. Instead, Meta continues to rely on victims to report images, as the Tazewell Sheriff’s Office did repeatedly before the suspect was arrested and charged. 


Even when the images are removed quickly, the damage is often already done. In the Virginia warrant, the targeted deputy said she’d suffered “significant emotional damage” because of the fakes and was concerned about her minor child seeing them. The Sheriff’s Office didn’t respond to requests for comment.


Meta told Forbes it had zero tolerance for AI nudifier apps and the imagery they created. They also said, however, that while it has detection technology to remove infringing content, the tools are not perfect, so it encourages people to report any photos they believe break its rules. 


This week, the Tech Transparency Project (TTP) released a report showing Meta’s tech can also fail to detect AI-generated CSAM when it was included in Facebook ads. TTP found 332 potentially infringing ads, most of which “showed a photo of a child, which was digitally altered with AI to make it appear the child was performing a sex act.” Some contained images of real children, according to the report, first covered by Wired


TTP director Katie Paul said that the findings, combined with the use of nudified images as harassment on Facebook, showed Meta wasn’t moving fast enough to deal with the issue. “This track record doesn’t inspire much confidence when it comes to Meta responding appropriately to AI-generated nude images on its platforms,” Paul told Forbes.


In response to the TTP report, Meta said it was trying to keep up with shifting tactics by those sharing AI-generated nudes. They noted that between November 2025 and January 2026, it removed over 344,000 ads on Facebook and Instagram that attempted to promote nudify apps. 


The issue is hitting all kinds of social media applications. In another warrant unearthed by Forbes this week, a Discord user was accused earlier this year of using AI to generate a video of a woman performing a sexual act that never happened. The video, shared via Discord private chat rather than a public thread, was generated with an image she’d shared with the suspect back in 2018 when she was a teenager.

Got a tip on surveillance or cybercrime? Get me on Signal at +1 929-512-7964.

Thomas Brewster Associate Editor, Cybersecurity

Follow me on Forbes.com

The NSA, CISA and the FBI issued a joint advisory warning that Chinese AI companies were effectively copying American AI models using a technique called “distillation” at scale. Distillation in this context happens when an entity or individual takes a large AI model and uses it to train a smaller, less power-hungry one. 


“China-based AI companies are engaging in aggressive, malicious, and targeted distillation activities at an industrial scale that extract restricted proprietary functionalities and capabilities of U.S. frontier AI models,” the federal agencies wrote.


China has denied it is supporting any such practice, according to NBC News, or that companies like DeepSeek or Alibaba are carrying out distillation.

The Stories You Have To Read Today

In the aftermath of OpenAI agents breaching Hugging Face, OpenAI called in some outside experts to carry out an investigation to supplement its own probe. But one of the contracted companies, METR, was not permitted to carry out research beyond the week in which the hack occurred and was only allowed access to OpenAI’s offices for a few days, per a New York Times report.


A federal judge has taken Anthropic’s side in its fight with the U.S. government, saying it unlawfully retaliated to the AI company’s decision not to sell its tech for use in mass surveillance when it labelled the business a “supply chain risk.” It’s yet to get any new contracts with the Pentagon since the public bust-up, however.


The White House has imposed tariffs of up to 100% on drones being imported into the U.S., which will largely affect Chinese manufacturers like DJI.

Winner Of The Week
Not quite a win, but a less bad loss: Hackers promised to return as much as 85% of the $320 million they stole from the Liquid Network, a Bitcoin sidechain. Sidechains connect to parent blockchains but can be given their own unique features that make them faster or more secure. The hackers say they’re keeping as much as $45 million of the stolen Bitcoin for themselves, though.
Loser Of The Week
A new identity theft service dubbed Nexus appeared on the dark web earlier this month selling digital scans of more than 153 million drivers licenses, pointing to a significant data breach affecting a massive number of Americans. The exact source of the breach is as-yet unconfirmed, but the FBI is investigating. Soon after it was reported by independent journalist Brian Krebs, Nexus was shut down without explanation.
More From Forbes
Most-Read From Forbes
Join Our Members-Only Event
Join Forbes journalists and higher education experts September 10 at 12 p.m. ET for an exclusive members-only discussion and live audience Q&A about choosing a college in the age of AI. Explore Forbes’ 2027 America’s Top Colleges list and learn what to consider when evaluating schools, majors, skills and the long-term value of a college education.
Subscribe To Register
Forbes

Unsubscribe from The Wiretap.

Manage Email Preferences

My Forbes Account  |  Newsletters  |  Help  |  Privacy

Forbes Media 499 Washington Blvd. Jersey City, NJ 07310

list_id:54119

Komentar

Postingan populer dari blog ini

πŸ’₯ Trending Games with potential token airdrops

🌟New Features Now Live on CoinCodex