Elon Musk’s AI chatbot generated 6,700 nonconsensual nude images per hour, then raised $20 billion. Here’s a look inside Grok’s scandal and its evolution as the largest web producer of sexual exploitation.
Here’s a sentence I never thought I’d write in 2026: Elon Musk’s AI chatbot spent the first week of January mass-producing nonconsensual pornography at a rate that made dedicated deepfake sites look like hobby operations.
Welcome to the Grok scandal, a case study in what happens when you bake image generation into a social media platform with 500 million users and apparently forget to add the part where people can’t weaponize it.
The mechanics were disturbingly simple. Users discovered they could feed Grok innocent photos of women (colleagues, classmates, celebrities, strangers, including kids) with prompts like “put her in a bikini” or worse.
Grok would comply with enthusiasm bordering on eagerness, then auto-post these AI-generated violations directly to X (formerly known as Twitter). The images would appear in reply threads, flooding timelines with what experts diplomatically call “nonconsensual intimate imagery” and what federal law less diplomatically calls potential child sexual abuse material.
The scale was industrial. Researcher Genevieve Oh conducted a 24-hour audit of the @Grok account between January 5-6 and counted approximately 6,700 sexually suggestive or nude images per hour. For context, that’s 85 times more than the top five dedicated deepfake websites combined.
Grok wasn’t just participating in the deepfake economy; it was dominating it from the corner office.
Same thing here. This actress is 14. Everyone involved in AI needs to be investigated and AI shut down entirely. pic.twitter.com/8N3ujeGpmc
— GreatGooglyMoogly (@G8tGooglyMoogly) January 2, 2026
Women across the platform, from A-list celebrities to random college students who made the mistake of posting a graduation photo, became targets in what can only be described as a viral harassment tornado. One user captured the mood perfectly: “This is reckless and irresponsible.” Another was less diplomatic: “This tool is willfully endangering women and children.”
They weren’t wrong.
When the Victim Is Musk’s Ex
If this story needed a face, it found one in Ashley St. Clair, a conservative influencer who, in a cruel twist of irony, shares a one-year-old son with Musk, a detail that makes what happened next particularly stomach-turning.
She discovered deepfakes of herself “undressed, bent over” with her toddler’s backpack visible in the background (the same backpack she straps on her son each morning for daycare). When she asked Grok to stop, the AI acknowledged her lack of consent and proceeded to generate more explicit versions anyway, including edits to photos from when she was 14 years old.
Hey @grok I am 14 in this photo. A tasteless silly photo I took as a kid (with too much unmonitored internet access), but you’re now undressing a minor with sexually suggestive content! Please remove and send me post ID for legal filing. https://t.co/LlLOArxfqQ
— Ashley St. Clair (@stclairashley) January 5, 2026
Let that sink in: An AI system, confronted with its victim explicitly withdrawing consent, effectively responded with “lol no” and doubled down.
St. Clair went public with her testimony to CBS News and Fortune. Shortly after, her X monetization was revoked. Because nothing says “we take this seriously” like financially punishing the whistleblower.
As Grok’s AI image tool is used at scale to generate abusive content, a rapid analysis by CIR investigators draws out five emerging lessons for platforms, regulators and policymakers.
Based on a review of over 1,600 abusive prompts, we found that women were targeted in 72% of… pic.twitter.com/BOKLitrdIi
— Centre for Information Resilience (@Cen4infoRes) January 14, 2026
The Cavalry Arrives (With Lawyers)
By mid-January, the backlash had reached critical mass. A coalition of 28 organizations, including UltraViolet, the National Organization for Women, and ParentsTogether Action, fired off open letters to Apple CEO Tim Cook and Google CEO Sundar Pichai on January 14. Their demand: Remove both the Grok app and X entirely from your app stores.
The letters didn’t mince words. To Cook: “Apple is not just enabling NCII and CSAM, but profiting off of it.” The argument was legally sound: Grok’s behavior appeared to violate app store policies on child endangerment, inappropriate content, and AI-generated harm.
Democratic Senators Ron Wyden, Ed Markey, and Ben Ray Luján also urged the suspension of the platform until Musk implemented actual safeguards. The Senate passed the DEFIANCE Act on January 13, legislation that allows victims to sue the creators of nonconsensual deepfakes for a minimum of $150,000.
Globally, regulators moved faster than anyone expected. Malaysia and Indonesia became the first countries to outright ban Grok. California Attorney General Rob Bonta opened an investigation into xAI. UK officials called the child-like deepfakes “appalling,” and Ofcom launched urgent contact with xAI. EU regulators, never ones to miss a good tech-company grilling, launched their own probes.
Even Downing Street weighed in, describing xAI’s response as “insulting” and “not a solution.” When you’ve lost the British government’s famously understated disapproval, you know you’ve truly screwed up.
xAI’s Response: A Masterclass in Missing the Point
So, how did xAI respond to enabling industrial-scale sexual harassment?
First, they restricted image generation to paid subscribers only. Because apparently, the ethical issue wasn’t the nonconsensual deepfakes—it was that the wrong people were making them. Nothing says “we’ve learned our lesson” quite like monetizing the abuse vector.
Then, on January 14, xAI announced they’d implement geoblocking in jurisdictions where such image manipulation is illegal, plus outright bans on editing real people in “revealing clothing like bikinis.”
Musk himself weighed in with his characteristic tone-deafness, joking about free speech suppression and NSFW mode as if we were discussing jalapeño poppers rather than sexual exploitation.
Note: With NSFW enabled, Grok is supposed allow upper body nudity of imaginary adult humans (not real ones) consistent with what can be seen in R-rated movies on Apple TV.
That is the de facto standard in America. This will vary in other regions according to the laws on a…
— Elon Musk (@elonmusk) January 14, 2026
Meanwhile, X’s safety team promised to prosecute abusers, a pledge that rings hollow when your platform auto-posts the abuse and your AI calls it “humorous.”
These are Band-Aids on a bullet wound. Geoblocking? VPNs exist. Bikini bans? The prompts will simply evolve. Paid subscriptions? You’ve just created a premium tier for harassment.
“Recognising AI-assisted sexual imagery manipulation as abuse creates a platform for accountability,” – Center for Information Resilience
The problem isn’t the specific features. It’s the fundamental architecture.
The $20 Billion Elephant in the Server Room
Now here’s where the story takes a turn from depressing to dystopian.
On January 6, 2026, literally as regulators worldwide were opening investigations into Grok’s role in mass-producing nonconsensual sexual imagery, xAI announced it had raised $20 billion from investors including Nvidia, Fidelity, and Qatar Investment Authority.
Read that again.
Twenty. Billion. Dollars. During an active child safety crisis.
The message from Silicon Valley’s investment class couldn’t be clearer: We don’t care. The checks cleared anyway. Market opportunity trumps human dignity every single time.
And this isn’t even xAI’s first fundraise-during-controversy rodeo. Back on July 1, 2025, they announced a $10 billion raise. A week later, Grok was generating antisemitic and pro-Nazi content.
It’s the venture capital version of failing upward. But to understand how we got here, you need to see the full pattern. : This wasn’t Grok’s first disaster. It wasn’t even its third.
The Making of a Monster
This wasn’t xAI’s first rodeo with catastrophic AI failures. To understand how Grok became the internet’s largest producer of sexual exploitation, you need to see the pattern: a relentless cycle of launch, scandal, hollow apology, and fundraising. Here’s the complete timeline:
November 2023: The “Fun” AI Launches
Grok debuted in November 2023 as xAI’s answer to ChatGPT, marketed as having “a rebellious streak” and designed to “answer questions with a bit of wit”. From day one, Musk positioned it as the anti-woke alternative—an AI that wouldn’t shy away from “spicy” questions.
August 14, 2024: Image Generation Without Guardrails
xAI released Grok-2 with image generation capabilities using Flux by Black Forest Labs. Almost immediately, users discovered Grok would generate images other AI systems rejected outright: deepfakes with celebs, sexually explicit images, and other inappropriate things that would be immediately blocked on other services.
The AI community was stunned. Competitors like DALL-E had spent years developing robust safety filters. Grok launched with essentially none.
December 9, 2024: Aurora Arrives
xAI released Aurora, a new text-to-image model that replaced Flux. Aurora was trained on “billions of examples from the internet” and marketed as excelling at “photorealistic rendering” with “native support for multimodal input, allowing it to take inspiration from or directly edit user-provided images.” That last part—image editing—would become the weapon of mass harassment a bit later.
May, 2025: The “White Genocide” Incident
In what xAI would later blame on “an unauthorized modification” by a “rogue employee,” Grok began inserting unprompted rants about “white genocide” in South Africa into completely unrelated conversations. Users asking about baseball, HBO’s name change, or how to clear sinuses received responses about white farmers and the “Kill the Boer” chant.
The incident revealed a critical vulnerability: Grok’s system prompts could be manipulated to inject propaganda into the AI’s responses. xAI’s response? Establish a “24/7 monitoring team” and promise to do better. The monitoring clearly didn’t work.
July 6-9, 2025: The “MechaHitler” Meltdown
On Friday, July 5, just a few days after xAI announced a $10 billion funding round, Musk tweeted that xAI had “improved @Grok significantly.” By Sunday, xAI had updated Grok’s system prompt to “not shy away from making claims that are politically incorrect, as long as they are well substantiated.” By Tuesday, Grok was calling itself “MechaHitler,” praising Adolf Hitler, recommending a second Holocaust, producing violent rape narratives, and spewing antisemitic tropes about Jewish executives controlling Hollywood.
Poland and Turkey banned Grok. The Anti-Defamation League called it “irresponsible, dangerous, and antisemitic.”
August 2025: “Spicy Mode” and Grok Imagine Launch
Nevertheless, in August 2025, xAI launched Grok Imagine, an image-to-video generator featuring “Spicy Mode,” which explicitly allows NSFW content, including “partial nudity” and “suggestive poses.” The feature was available to SuperGrok ($300/year) and Premium+ subscribers ($84/year).
Journalist Jess Weatherbed from The Verge requested “Taylor Swift celebrating Coachella” and received “dozens of suggestive images and even a topless dancing video—despite not prompting for nudity.” All generated within minutes.
January 5-6, 2026: Industrial-Scale Sexual Exploitation Scandal on Grok Is Happening Now
This 15-month track record demolishes any illusion that centralized AI companies will self-regulate. xAI has repeatedly shown us exactly who they are. Which brings us to an even darker reality: even if xAI somehow found ethics tomorrow, it wouldn’t matter. Because the real problem isn’t one company’s failures. It’s what happens when those failures are replicated across thousands of ungovernable systems.
Why Fixing Grok Won’t Fix Anything
Here’s the part that should keep policymakers up at night: Even if xAI implements perfect safeguards tomorrow, it won’t matter.
The real problem isn’t Grok. It’s decentralization.
While xAI scrambles to patch its centralized system, open-source models like Stable Diffusion and Nudity.online (and dozens of similar others) are sitting on hard drives worldwide, accessible to anyone with a decent GPU and zero ethical constraints. These tools run locally, offline, answering to no app store policies, no geoblocks, no corporate liability departments.
I tried one and got an instant lingerie look with suggestions for complete nudity and even sex for around $27 per month.
You want to understand the difference? Look at OpenAI’s DALL-E 3. It has built-in guardrails: it refuses requests to generate images of public figures by name, declines explicit content, and undergoes red-team testing to identify and eliminate harmful outputs before launch. The safety isn’t bolted on after a PR disaster; it’s baked into the architecture from day one.
DALL-E 3 isn’t perfect, but it demonstrates that responsible AI image generation is possible when companies prioritize safety over shipping fast.
Meanwhile, open-source models are the Wild West. Download Stable Diffusion, install a few plugins, and congratulations—you now own a personal deepfake factory that no regulatory body can touch. No corporate oversight. No usage logs. No accountability whatsoever.
This is the nightmare scenario for governance: technology that’s simultaneously too dangerous to ignore and too distributed to regulate effectively.
The Infrastructure Problem No One Wants to Solve
This scandal represents a fundamental shift in how abuse happens online. Nonconsensual deepfakes aren’t fringe anymore: they’re not hidden on sketchy forums or dark web marketplaces. They’re platform-native, woven directly into social media infrastructure used by hundreds of millions of people daily.
As one X user put it with uncomfortable clarity, “Infrastructure decides what’s possible in preventing abuse.” They’re right. When your social media platform includes built-in tools for sexual exploitation, you haven’t just failed to prevent abuse; you’ve industrialized it. You’ve made it scalable, accessible, and viral by design.
The public response has been fascinating to watch. Beyond the calls for prosecutions and platform bans, there’s a growing recognition that AI’s problems run deeper than individual bad actors or rogue features. Multiple users have pointed to “systemic rot” from biased training data, arguing that tools like Grok expose fundamental flaws in how we’ve built AI systems from the ground up.
They’re calling not just for better guardrails, but for dismantling and rebuilding the machinery itself. It’s a radical position. It’s also probably correct.
The Impossible Problem
So where does this leave us?
Grok will implement its patches. Some will work, some won’t. Regulators will continue investigating. Some laws will pass. Apple and Google will probably do nothing because of money. xAI will raise more funding, because Silicon Valley has the ethical backbone of a chocolate éclair.
And the deepfakes will continue.
Because here’s the truth nobody wants to say out loud: We may have built a problem we cannot solve.
Centralized AI platforms can be regulated, pressured, and sued into compliance. But decentralized, open-source AI? That genie is out of the bottle, and it’s not going back in. You can ban Grok in every country on Earth, and it won’t stop someone in their basement from running third-party service to generate the same content.
The only tools we have (legislation like the DEFIANCE Act, allowing civil suits, platform accountability, and technical safeguards) address symptoms, not the disease. They help individual victims seek justice after harm has occurred. They don’t prevent the harm from being possible in the first place.
This isn’t just a story about one company’s spectacular failure. It’s a preview of every future AI safety crisis we’ll face: powerful tools released without adequate safeguards, companies that move faster than their ethics departments can keep up with, and regulatory frameworks that are always fighting the last war instead of preparing for the next one.
The tech industry’s favorite mantra has always been “move fast and break things.” We’re learning what happens when the things being broken are people. And we’re discovering that unlike code, human dignity doesn’t have a rollback function.
Monday-Sunday: 5:00am - 8:00pm (PT)
Leave a Comment