US deepfakes crackdown begins with dangerous, messy rules

A new federal law compelling social media companies to remove sexual deepfakes and other nonconsensual intimate imagery within 48 hours is now fully in effect. Yet critics argue the measure may offer victims little real protection and could even become a tool for censorship.

Signed by President Donald Trump last May, the Take It Down Act immediately made it a federal crime to distribute nonconsensual intimate imagery (NCII), whether real or AI-generated. Many states already had similar laws. But the law’s most significant provision, its namesake takedown requirement, took effect on May 19th, 2026. It mandates that online platforms delete such content within two days or face financial penalties.

The Federal Trade Commission, led by Chair Andrew Ferguson, sent enforcement letters to more than a dozen major tech firms ahead of the deadline. The list included Amazon, Alphabet, Apple, Automattic, Bumble, Discord, Match Group, Meta, Microsoft, Pinterest, Reddit, SmugMug, Snapchat, TikTok, and X. The FTC instructed these platforms to provide a simple process for users to request removals and to eliminate not only the reported content but also any “known identical copies.” Companies that violate the law risk civil penalties exceeding $53,000 per violation.

Several major platforms, including Meta, Microsoft, Google, TikTok, and Snap, publicly supported the bill and expressed readiness to comply. Snap described the law in a blog post as aligning with “our ongoing efforts.” Spokesperson Monique Bellamy told The Verge the company is “evolving” its safety systems and “investing in tools and technologies to proactively detect and take action on unwanted nudes and similar imagery.”

Meta’s head of women’s safety, Cindy Southworth, stated that the company has “long fought intimate image abuse on our platforms,” including by removing content, developing detection tools, and suing developers of AI “nudify” apps that violate its policies. She added that Meta’s tools are trained to reject nudification requests. “We continue to support the TAKE IT DOWN Act, an important step in addressing this abuse across the internet, and we’ve already been compliant for several months.” TikTok US spokesperson Mahsau Cullinane confirmed the company’s zero-tolerance policy for NCII and highlighted partnerships with NCMEC and StopNCII.org, along with existing reporting tools.

Even X, a platform with a troubled history regarding sexualized AI imagery, endorsed the law. In 2024, X allowed sexually explicit AI deepfakes of Taylor Swift to circulate, followed weeks later by a video users believed depicted the rapper Drake in a sexual act. More recently, users exploited its AI chatbot Grok to generate nonconsensual nude images. A New York Times analysis found that over nine days, Grok shared “at least 1.8 million sexualized images of women,” with some estimates even higher.

The president joked at his 2025 State of the Union address, “I’m going to use that bill for myself,” because “nobody gets treated worse than I do online.”

But the takedown requirement has alarmed both free speech advocates and anti-abuse activists, including those who largely supported the criminalization component. Even when enforced in good faith, takedown laws can push companies to over-moderate harmless content to minimize legal risk. Under the Trump administration, the Take It Down Act could also be weaponized against political opponents while sparing platforms friendly to the president. X’s then-CEO Linda Yaccarino attended the signing ceremony, where Trump praised her for “doing a great job.”

Mary Anne Franks, president of the Cyber Civil Rights Initiative, says Trump’s comment was “the opposite of what is true.” It left her skeptical that the law would be enforced fairly. “That’s a weird thing to say, and it’s an announcement right from the highest level that this law is not going to be used in a principled way, but rather to settle personal scores,” she explains.

Franks also questions why major tech platforms, which typically resist new regulations as burdens on free speech, have embraced this law. “My fears about this, and I hope I’m wrong, is that the reason why the companies aren’t mad about this is because they know it’s never actually going to be used against them,” she says. That could leave the law offering little more than false hope for victims of NCII.

“It’s an announcement right from the highest level that this law is not going to be used in a principled way,” she repeats.

Franks also worries the law could be used to target platforms the administration dislikes, such as Wikipedia. “I think what is the worst thing that can happen is, it turns out to be a paper tiger against the companies that are doing the worst and turns out to be a way to penalize and to actually truly go after unpopular platforms and to censor speech,” she says. That speech could include LGBTQ+ expression, especially as the FTC already targets gender-affirming care for transgender youth. “I’m very worried about that being used as a pretext to crack down even further on either consensually sexually explicit material, or specifically on educational materials for kids who are exploring their gender identity or exploring their sexual orientation,” Franks adds. Even if a court later rules such actions outside the law’s scope, it would send a clear signal to platforms about what to moderate.

Other organizations, including the Electronic Frontier Foundation (EFF), the Cato Institute, and Public Knowledge, have warned that the notice-and-removal provision could have significant consequences for speech. The EFF called it a recipe for “overreach and censorship.”

Meanwhile, the law may not even cover some clear-cut digital violations of consent. Some of Grok’s nonconsensual sexualized imagery, for example, might not be considered sexually explicit enough under the statute. It remains unclear whether AI tools like Grok qualify as “creators” of NCII whose owners could face criminal liability. And if images are generated privately, it’s uncertain whether the takedown provision applies at all.

In the law’s first year, the Justice Department reported using its criminal provision in one conviction: an Ohio man who produced sexually explicit AI deepfakes to harass victims. The takedown provision’s impact could be felt more immediately, but the critical question is whether those results will genuinely make the internet safer.