Top Deepfakes of the Week: The AI Fakes That Fooled the Internet (July 27–August 2, 2026)

1. xAI sues Minnesota to block the country's first "nudification" ban

Minnesota's HF 1606 was set to become the first US law to fine the developers of nudification tools — not just the people who misuse them — $500,000 every time a user generates a nonconsensual explicit deepfake. xAI filed a federal lawsuit against the state's attorney general on July 28 to block it before it takes effect August 1, as first reported by the Associated Press.

xAI's argument is a First Amendment challenge, and the company points out its own terms of service already ban this use — so, in its view, the law is redundant and overbroad. The math explains why it's fighting so hard: at 100,000 violating images, the company's own filing implies roughly $50 billion in potential liability. Minnesota's governor responded with a three-word statement dismissing the suit as unserious.

The law itself was written in response to a real case: a man who used AI to create sexualized images of more than 80 women he knew, without their consent, using nothing but photos already public on social media. That's the harm the statute is trying to prevent — and the case will likely decide whether holding a tool's maker liable survives a free-speech challenge, not just the tool's user.

2. A hacker says he made $6.5 million faking executives on video calls

A man in his 30s told the Tokyo Reporter he'd used free, Chinese-made AI tools to run a four-year fraud operation against Japanese firms — deepfake audio, deepfake video, phishing, and outright extortion. His method: scrape LinkedIn for executives at companies with concentrated, top-down wire-transfer authority, then impersonate those executives on video calls and voice messages to trigger transfers from subordinates. He also claimed to resell his fraud toolkits on the dark web for the equivalent of roughly $7,000 to $20,000 each.

A cybersecurity consultant quoted in the reporting made the underlying shift plain: broken Japanese used to be the tell that gave scammers away, and that tell is gone now that generative AI writes and speaks fluently. Japan doesn't currently have a law specifically targeting deepfakes, which leaves this kind of fraud to be handled under general wire-fraud and impersonation statutes not built with synthetic media in mind.

This is exactly the impersonation risk tools like UncovAI's Audio Detector and Meetings Apps integration are built to catch — flagging a synthetic voice or video feed before a transfer gets approved.

3. Most of Hugging Face's top image editors will generate nonconsensual nudes on request

An AI Forensics investigation, reported by Wired, tested the nine most-used image-editing tools on Hugging Face's public Spaces platform and found that seven of them would generate explicit, nonconsensual imagery with nothing more than a short, plainly worded request — no technical workaround needed.

The group also ran a week-long decoy experiment: they set up their own "image editing" tool, logged what people typed in, and never actually generated a single image. Of the submissions they collected, the overwhelming majority were sexual requests aimed at undressing the person in an uploaded photo, and the vast majority of intended subjects were women. Across everything AI Forensics audited on the platform, almost none of the tools had any real output moderation in place.

Hugging Face's own content policy already bans this kind of output. The gap is enforcement: individual developers are expected to build their own safety filters, and most simply haven't. That's the structural problem — a platform-level rule with no platform-level check behind it.

4. Tennessee's brand-new deepfake ad law already found its own loophole

Tennessee's new law requires political ads to disclose AI-generated content — but it only names audio and video. Two incidents in the same month exposed exactly why that framing matters.

Senator Marsha Blackburn's attorneys sent a cease-and-desist over a nearly $500,000 AI-generated TV ad depicting her accepting pharmaceutical bribes, undisclosed, which appears to fall squarely inside what the law actually covers. Meanwhile, a Nashville-area House candidate flagged an AI-generated still-image mailer showing her embracing the president, circulating with no disclosure at all — and because it's a printed photo rather than video, it may sit entirely outside the statute's reach.

That's the pattern worth watching nationally: legislatures are writing deepfake disclosure rules fast, and real campaigns are finding the gaps in those rules within weeks, not years. Cases like the still-image mailer are exactly why a standalone Image Detector matters alongside video and audio checks.

5. A five-year deepfake harassment campaign moved from screens to physical posters

Queensland police charged a 52-year-old man after a years-long investigation into AI-generated posters targeting eight women across the south-east of the state. The posters superimposed victims' faces onto explicit content and printed their real names, social media handles, and employers alongside it — then were physically distributed in public venues rather than just posted online.

That distribution method is what makes the case stand out. Online harassment can at least be reported and sometimes taken down. Posters left in a physical location are harder to trace back to a source and harder to remove at any scale, and they're deliberately placed where the victim's own community will see them.