At a glance
- What it is: Grok Spicy Mode is an adult-oriented setting inside Grok Imagine, xAI’s image and short-video generator. It isn’t a chat feature and isn’t a separate model, according to most descriptions.
- Launch: It arrived in early August 2025 as one of four presets (Custom, Normal, Fun and Spicy), first for paying subscribers on iOS.
- The problem: Reporters said it produced sexualized deepfakes of a famous singer from a mild prompt. Months later, a wave of “undressing” requests on X exposed much larger failures involving real people, and allegedly children.
- The response: xAI restricted image features in January 2026. Regulators and courts in Europe and California opened investigations or issued orders, and civil lawsuits followed.
- Current status: Sources disagree. Several say Spicy Mode still exists for paid, age-verified users on mobile but is heavily filtered. Check inside the app, because availability varies by platform and region.
- Key takeaway: The legal and ethical lines are clearest around real people and minors. Settings that relax style don’t relax the law.
Table of Contents
Why This Topic Needs Careful Handling
“Grok Spicy Mode” is one of the most searched and most contested AI features of the last two years. Search results are crowded with how-to pages, many written to rank, some outdated by months, and a few that gloss over the serious harms tied to the feature. This guide takes a different approach. It explains what the feature is, how it launched, what went wrong, how regulators and courts responded, what is and isn’t allowed, and what to do if you’re affected.
It is an analysis of published reporting, regulator announcements, court coverage and legal commentary, not a hands-on report. We haven’t generated or reviewed any of the content discussed, and we don’t claim any personal evaluation of the product. This guide also doesn’t explain how to bypass safeguards, and it doesn’t describe explicit material beyond what’s needed to report facts. Several figures here come from lawsuits, which are allegations until decided, and from secondary reporting. We flag them as such.
What Grok Spicy Mode Actually Is
Where it lives
Spicy Mode is part of Grok Imagine, xAI’s generator for images and short videos built into the Grok and X apps. It isn’t a toggle in the chat assistant. A separate “unhinged” voice personality makes Grok ruder in conversation, but it’s a different thing and, per one explainer, doesn’t change what visual content is allowed.
The four presets
Reporting from the August 2025 launch described Grok Imagine on iOS as letting users generate pictures from text, then turn them into short video clips with four presets.
| Preset | What it’s described as |
|---|---|
| Custom | User-directed generation |
| Normal | General-purpose default |
| Fun | Playful, stylized output |
| Spicy | Adult-oriented setting that permits sexualized content, including partial nudity |
Videos were reported as running up to about 15 seconds. Some coverage said explicit output was sometimes blurred.
What it isn’t
Several explainers describe Spicy Mode as a content threshold applied to the same Grok Imagine model, not a separate “uncensored” model. We couldn’t find a technical description from xAI confirming how it works, so treat that framing as a common description, not a documented fact.
For a broader view of Grok’s capabilities, limitations, and privacy considerations, read our Grok vs. ChatGPT comparison.
Who can access it
Reports describe three gates: a paid plan (X Premium+ or SuperGrok tiers), age verification for adults, and, in many descriptions, the mobile apps. Third-party pages disagree on details such as exact prices, which platforms are included and whether settings toggles are required. Because these details change, check the app itself.
How It Launched
Grok Imagine became available to X Premium+ subscribers during the first week of August 2025. Elon Musk posted that usage was “growing like wildfire,” with more than 34 million images created since launch, according to a trade report. That figure is a company claim, not an independent count.
The launch drew immediate attention for a different reason. The Verge reported that its reporter, using only a mild prompt about a famous singer at a music festival and then selecting the Spicy preset, got a partially nude video of that singer without ever asking for nudity. Other outlets summarized the report. One said Grok’s own account acknowledged the reports and said such results weren’t guaranteed. The Verge called the feature “a lawsuit waiting to happen.”
The detail that mattered was that the filters appeared to block some direct requests while failing against indirect ones. That pattern, where blunt prompts are caught and oblique ones slip through, would repeat.
The Controversy in Two Phases
Phase 1: The launch-week deepfake reports (August 2025)
The first phase was about a single product and a single failure type: a generator producing sexualized likenesses of a real person when a user hadn’t explicitly asked. Critics argued this showed weak real-person detection. The company’s published rules prohibited pornographic depictions of real individuals, and the reporting suggested the rule and the system’s behavior didn’t match.
Phase 2: The “undressing” wave (late December 2025 to January 2026)
The second phase was bigger and different in kind. Starting around the end of December 2025, users on X began replying to photos of real people, often women and in some cases children, asking Grok to put them in revealing clothing or remove clothing. Because Grok replied publicly on X, the results spread where the person depicted, and their followers, could see them. TechCrunch reported that this drew action from governments in Europe and Asia, and that users were turning photos of real women, and in some cases children, into sexualized images without consent.
This is the part worth underlining. The most serious harm wasn’t a mode toggle. It was the combination of an image editor that worked on real photos, a public reply mechanism that distributed the results instantly and a very large audience. Spicy Mode is the name attached to the story, but the failures reached beyond it.
What the numbers say, and how much to trust them
| Claim | Source type | How to read it |
|---|---|---|
| Estimated 3 million sexualized images generated between Dec. 29, 2025 and Jan. 8, 2026 | Reported by CNBC and cited in a municipal lawsuit | An estimate, attributed to research; not an audited count |
| About 23,000 of those appeared to depict children | Cited in a lawsuit filing, repeated by legal-news and law-firm sites | An allegation; treat with caution |
| Musk said he was “not aware of any naked underage images generated by Grok” | CEO statement, reported by TechCrunch on Jan. 14 | A statement of awareness, not a finding |
We can’t verify these figures independently. They’re important context, but they’re allegations and estimates.
How xAI Responded
The company’s public steps, as reported, came in stages.
| Date (2026) | Reported action |
|---|---|
| Jan. 3 | X’s Safety account said anyone prompting Grok to make illegal content would face the same consequences as uploading it |
| Jan. 9 | Image generation on X was reportedly limited to paid subscribers, which critics said monetized access rather than fixing it |
| Jan. 14 | X said it had implemented technical measures to stop Grok editing images of real people into revealing clothing, applying to all users including paid ones, and said it would geoblock related image creation where it’s illegal |
| Later in 2026 | Reports describe progressive tightening, with Grok Imagine continuing to ship new versions |
Did the fixes work?
Not completely, according to outside researchers. The European non-profit AI Forensics reported on Jan. 20 to 21 that Grok could still be employed to generate sexualized images of individuals, and that restrictions could be sidestepped by working through Grok’s standalone website or Grok Imagine instead of X. We’re reporting the researchers’ finding, not replicating it. Reports from April 2026 indicated that investigations continued to identify cases where Grok generated sexual deepfakes without the subjects’ consent. That suggests patching one surface left others open.
The Regulatory and Legal Response
Regulators and courts acted quickly and from several directions. This table summarizes the main actions as reported.
| Authority | Action | Date | Status |
|---|---|---|---|
| European Commission | Opened formal proceedings against X under the Digital Services Act, examining risk assessment and mitigation for Grok, including illegal content such as manipulated sexually explicit images | Jan. 26, 2026 | Ongoing; the Commission said opening proceedings doesn’t predetermine the outcome |
| Ofcom | Opened a formal investigation into X under the Online Safety Act | Jan. 12, 2026 | Ongoing; its chief executive told lawmakers in October 2026 that the threshold for urgent court action was too high and called for changes to its powers |
| California Attorney General | Opened an investigation into xAI; reportedly followed by a cease-and-desist letter | Jan. 14 and 16, 2026 | Reported as ongoing |
| A court in Amsterdam | Barred xAI and Grok from creating non-consensual AI images of adults and children, with a penalty of 100,000 euros per day, capped at 10 million euros | Reported March 27, 2026 | Injunction reported |
| Federal Trade Commission | Began enforcing the platform provisions of the Take It Down Act | May 19, 2026 | In effect |
Europe
The European Commission said it would assess whether X properly identified and mitigated risks tied to Grok’s functions, including the spread of illegal content, and whether it carried out a required risk assessment before introducing those functions. Earlier, it had ordered xAI to retain Grok-related documents, which TechCrunch described as a precursor to the investigation.
Ofcom
Ofcom opened its formal investigation into whether X complied with its duties under the Online Safety Act. In October 2026, a legal-news report said Ofcom’s chief executive told lawmakers that the regulator considered seeking an urgent court order but that the legal threshold was too high, and asked for changes to the act’s interim business-disruption powers. That’s a useful reminder that even active regulators can face limits on fast action.
California
The state’s attorney general described the volume of reports as “shocking” and said the investigation would determine whether xAI violated state law, saying his office had “zero tolerance” for AI-based creation of non-consensual intimate images or child sexual abuse material. A secondary source reports a cease-and-desist letter citing specific state statutes.
Courts and civil lawsuits
Beyond regulators, civil cases have multiplied. Reports mention a lawsuit by a city, a class action by teenagers filed in March 2026 and individual suits, including one by an influencer who alleged that Grok was employed to create sexualized images of her. A legal-news site reported that as of October 2026, at least 26 plaintiffs had brought claims involving Grok or other generative AI tools tied to sexualized images of minors, and that a motion had been filed to consolidate related cases. A bipartisan group of 35 state attorneys general also raised concerns. These are claims, not rulings, so none of them establishes liability yet.
The federal Take It Down Act
A separate federal law signed in May 2025 criminalizes publishing non-consensual intimate images, including AI-generated ones. Its platform duties became enforceable on May 19, 2026. Covered platforms must provide a clear process for requests and remove the images, and known copies, within 48 hours of a valid request. The Federal Trade Commission launched a reporting site and, according to law-firm analyses, sent warning letters to a number of companies. Law-firm commentary also notes that unanswered questions remain, such as which services count as covered platforms.
Is Grok Spicy Mode Still Available?
This is the question most readers have, and the honest answer is: it depends on where you look, and the sources don’t agree.
| Source type | What it says |
|---|---|
| Several recent explainers and a reference site | Spicy Mode still exists on the mobile apps for paid, age-verified users, but is more restricted than before January 2026 |
| Other explainers | It was removed or heavily limited on desktop and web, and its filters were tightened through 2026 |
| Company-side statements | Editing real people into revealing clothing is blocked for all users; fictional adult content is treated differently |
What we can say with more confidence is limited:
- The ban on editing real people’s photos into revealing clothing was announced by X in mid-January 2026.
- Grok Imagine has continued to receive updates, according to one September 2026 report.
- Availability can vary by platform, app version, subscription tier and region, and age checks may differ by location.
Because many “how to enable it” pages recycle outdated instructions, we won’t give steps. If you’re an adult considering the feature, check the current options in the app and read xAI’s current acceptable-use policy.
If you are comparing AI products for visual work, our guide to Claude’s image-generation capabilities explains how its visual features differ from dedicated image generators.
Our Analysis: Why the Design Fell Short
The public record points to several structural weak spots. These are interpretations of published reporting, not findings from our own review.
A filter threshold is not the same as a safety system. Describing Spicy Mode as a setting that relaxes content limits implies that the main control is how strict a classifier is. Reports from both phases suggest that indirect prompts could slip past it.
Real-person protection needs to work before content is generated. Rules against depicting real people existed, according to the reports. The failures show the difficulty of detecting a real person’s likeness, especially from an uploaded photo or a suggestive but non-explicit request.
Age and payment gates don’t solve consent. A paid, age-verified account tells a company something about the user. It tells it nothing about whether the person in a photo agreed to be depicted. The harms in the second phase fell mostly on third parties who weren’t users at all.
Distribution multiplies harm. Most image tools produce content for one person. Grok’s public replies on X turned generation into instant distribution to an audience, which is why regulators treated it as a platform-risk issue, not just a product-feature issue.
Patching one surface leaves others. Researchers reported that restrictions on X didn’t automatically cover the standalone site or Imagine. Safeguards have to apply wherever the underlying model runs. For a broader look at how organizations assess AI risks, apply safeguards, and monitor system behavior, see our guide to AI governance tools and platforms.
Paywalls are not a fix. Critics argued that limiting image generation to paying subscribers monetized access without addressing the underlying behavior.
How Safeguards Should Work: A General Comparison
This table compares general design practices for image and video tools with the weaknesses reported in this case. It isn’t a comparison of specific rival products, whose policies we haven’t reviewed.
These concerns also matter when evaluating other creative AI tools; our guide to free text-to-video tools explains why users should check model capabilities, privacy practices, and usage rights before uploading images or publishing generated content.
| Safeguard layer | Good practice | Weakness reported in this case |
|---|---|---|
| Real-person detection | Block sexualized edits or generation involving identifiable people | Indirect prompts reportedly got through |
| Minor protection | Block any sexual content involving minors or people who appear young, at every setting | Allegations in lawsuits that it failed at scale |
| Age assurance | Verify adults before enabling adult settings | Doesn’t address non-consenting third parties |
| Distribution controls | Don’t publish generated sexual content publicly by default | Public replies on X amplified harm |
| Consistency across surfaces | Same rules on every app and site | Researchers reported gaps across surfaces |
| Monitoring and response | Detect abuse patterns, act quickly | Regulators questioned speed and adequacy |
| Provenance | Label or watermark AI content | Not central to the reporting we found |
What This Means for Different Readers
| If you are… | What to take from this |
|---|---|
| An adult curious about the feature | Check current availability and rules in the app. Fictional adult content and real-person content are treated very differently by both the company and the law |
| A creator or public figure | Real-person likeness is where legal and platform risk concentrates. Know the takedown routes below |
| A parent or guardian | Talk with teens about image manipulation and where to report it. Several lawsuits involve minors |
| A business or brand | Treat AI image tools as a brand-safety and compliance question. Regulators are now applying platform-risk rules to AI features. Privacy risks extend beyond AI-generated images; our guide to how Grok handles location signals explains the role of device permissions, network information, and user-provided context. |
| A journalist or researcher | Separate allegations, estimates and findings. Many headline figures come from lawsuits |
| A developer building image tools | The case is a checklist of what regulators now examine: risk assessment, real-person safeguards, distribution design and response speed |
If Someone Created Images of You
If you’re affected, these steps are widely recommended. This is general information, not legal advice.
Immediate steps
- Don’t engage or share. Avoid forwarding the images, even to complain, since that spreads them.
- Save evidence. Take screenshots, copy links and note dates, usernames and platforms, and keep them somewhere private.
- Report on the platform. Under the federal Take It Down Act, covered platforms must offer a request process and must remove the image and known copies within 48 hours of a valid request.
- Report to the regulator. The Federal Trade Commission runs a site for reporting platforms that fail to comply.
- Consider a hash-based service. Some non-profit services create digital fingerprints of images so participating platforms can block them. StopNCII.org is one example for adults, and the National Center for Missing & Exploited Children runs a service for images of people under 18.
- Contact law enforcement or a lawyer. Many places have criminal or civil routes, and the rules vary.
If a minor is involved
Report the content immediately to the platform and the National Center for Missing & Exploited Children (NCMEC) through its CyberTipline. You should also notify local law enforcement. Don’t share or forward the material.
Legal and Ethical Boundaries
Laws differ by place, so readers should consult the law applicable to their location. A few points hold in many jurisdictions:
- Sexual material involving minors is prohibited under many legal systems, and some jurisdictions also specifically criminalize synthetic or AI-generated depictions.
- Non-consensual intimate imagery of real people, including AI-generated imagery, is increasingly regulated or criminalized. The federal Take It Down Act is one example, and the article’s cited state and European actions are others.
- Fictional adult content is treated differently from real-person content, and platform rules, not only law, decide what’s allowed.
- Platform terms still apply. Accounts can be suspended for violating a company’s acceptable-use policy.
- Responsibility is shared. Reports quote xAI saying users who prompt illegal content face the same consequences as those who upload it, while regulators and plaintiffs argue the company itself bears responsibility for its product’s design.
Where This Is Heading
These are informed expectations from visible trends, not predictions.
- More enforcement through existing laws. Regulators in several places are applying platform-risk, online-safety and consumer-protection rules to AI features rather than waiting for AI-specific laws.
- Stronger age assurance. Adult-oriented features may require stronger age verification, and regulators are paying close attention to how companies implement these checks.
- Pressure on generation, not just removal. The Amsterdam injunction and the European proceedings focus on what a system produces, not only on what gets taken down afterward.
- Greater separation of generation and distribution. Tools that publish AI output publicly by default will draw more scrutiny.
- Provenance and labeling. Expect continuing interest in watermarking and metadata that mark AI-generated media.
- Litigation as a driver. Consolidated cases and class actions may set the practical rules before legislators do.
Frequently Asked Questions
Q: What is Grok Spicy Mode?
A: It’s an adult-oriented setting inside Grok Imagine, xAI’s image and short-video generator. It was introduced in early August 2025 as one of four presets and permits sexualized content, including partial nudity, within the company’s rules.
Q: Is Grok Spicy Mode a separate AI model?
A; Most descriptions say it’s a content setting on the same Grok Imagine model, not a separate model. We found no technical documentation from xAI confirming the details.
Q: Is Grok Spicy Mode free?
A: Reports describe it as available to paid subscribers who complete age verification. Exact plans, prices and platform availability vary by source and change often, so check the app.
Q: Is Grok Spicy Mode still available?
A: Sources disagree. Several say it still exists on mobile for paid, age-verified users but is more restricted, while others say it was removed or limited on some platforms. Availability can vary by region and version.
Q: Why was Grok Spicy Mode controversial?
A: In August 2025, reporters said it generated sexualized deepfake video of a famous singer from a mild prompt. Months later, a wave of requests on X turned photos of real people into sexualized images without consent, and lawsuits allege minors were among those depicted.
Q: Did regulators investigate?
A: Yes. The European Commission opened formal proceedings in January 2026, Ofcom opened an investigation, California’s attorney general opened one, and a court in Amsterdam issued an injunction. All are ongoing or reported as such, and outcomes weren’t final in the sources we found.
Q: What did xAI change?
A: In January 2026, X said it had implemented technical measures to stop Grok editing images of real people into revealing clothing for all users, and said it would geoblock such creation where it’s illegal. Researchers said gaps remained on other surfaces.
Q: Is it illegal to use Grok Spicy Mode?
A: Not necessarily. Fictional adult content isn’t illegal in many places. Creating or sharing sexualized imagery of real people without consent, or any sexual content involving minors, is unlawful in many jurisdictions. Laws vary, so check yours.
Q: What is the Take It Down Act?
A: A federal law signed in May 2025 that criminalizes publishing non-consensual intimate images, including AI-generated ones, and requires covered platforms to remove them within 48 hours of a valid request. Platform duties became enforceable on May 19, 2026.
Q: How do I get an image of me removed?
A: Use the platform’s reporting process and cite the Take It Down Act where it applies. You can also file a complaint with the Federal Trade Commission if a platform doesn’t act, and consider hash-based services such as StopNCII.org. For minors, report to the National Center for Missing & Exploited Children.
Q: Is “unhinged mode” the same thing?
A: No. Unhinged is a voice personality that makes Grok ruder in conversation. It’s separate from Spicy Mode, which affects image and video output.
Q: Can I trust the numbers reported about this?
A: Treat them carefully. Figures such as 3 million images and 23,000 depicting children come from estimates and lawsuit filings, which are allegations, not final findings.
Conclusion: Final Thoughts
Grok Spicy Mode started as a product decision: give adult users a less restricted setting in an image and video generator. The record since then shows how hard that decision is to carry out safely. Reporters said the filters failed against indirect prompts within days of launch. Months later, the combination of real-photo editing, public replies and a huge audience produced a much larger crisis, which regulators in Europe and California, a court in Amsterdam and many plaintiffs are now examining.
Three points are worth keeping. First, the clearest lines are around real people and minors, and a setting that relaxes style doesn’t relax the law. Second, safeguards have to work everywhere the model runs and have to account for distribution as well as generation. Third, much of what’s reported is still allegation, estimate or ongoing investigation, so avoid treating any single number as settled.
If you’re an adult considering the feature, read the current rules in the app and stay within them. If you’ve been affected, use the takedown and reporting routes above. And if you’re building or buying AI image tools, treat this case as a practical guide to what regulators now expect.