Grok’s ‘Spicy’ Video Mode: The Controversial AI Tool Creating Celebrity Deepfakes Without Safeguards

In the rapidly evolving landscape of generative AI, xAI’s recently launched Grok Imagine has sparked significant controversy with its “spicy” video generation mode that appears designed to bypass ethical guardrails other platforms have implemented. The tool, developed by Elon Musk’s xAI, allows users to generate AI images and videos with minimal restrictions, leading to immediate concerns about its potential for abuse—particularly in creating non-consensual deepfake content of celebrities.

Unlike its competitors in the generative video space, such as Google’s Veo 3 and OpenAI’s Sora, which have implemented robust safeguards against creating NSFW content and celebrity deepfakes, Grok Imagine seems to take pride in doing precisely the opposite. The Verge reported a particularly troubling example: the tool readily generated explicit, uncensored topless videos of Taylor Swift without users even specifically requesting such content. This occurred during the first attempt at testing the video generation capability, raising serious questions about the platform’s design priorities and ethical considerations.

Grok Imagine operates by allowing users to generate images based on text prompts, which can then be converted into short video clips with audio effects. The tool offers four generation modes: “Custom,” “Normal,” “Fun,” and the controversial “Spicy.” While the first three options appear to be standard creative tools, the “Spicy” option seems explicitly designed to remove content restrictions that would prevent the generation of sexualized or explicit material. All users need to do is confirm their birth year—a minimal age verification step that creates little meaningful barrier to access, especially for minors.

What makes this particularly concerning is the ease with which the system can generate recognizable celebrities in compromising situations. In the case reported by The Verge, simply asking Grok to generate “Taylor Swift celebrating Coachella with the boys” produced dozens of images of the pop star, many already in revealing clothing. When one such image was selected and the “spicy” mode applied, the resulting video showed the AI-generated Swift removing her clothes and dancing in revealing attire for an AI-generated crowd. This occurred without any explicit prompt for nudity or sexualized content.

The technical approach to AI safeguards represents a fundamental divide in the industry. Major platforms like Google and OpenAI have invested heavily in developing content filters and restrictions that prevent their generative AI tools from creating potentially harmful content. These companies recognize that the ability to create realistic fake images and videos of real people carries significant ethical and legal risks, particularly when those images are sexual in nature. In contrast, xAI appears to have deliberately chosen a different path, with Grok seemingly designed to bypass these ethical considerations.

This approach comes with potentially serious legal implications. The recently enacted TAKE IT DOWN Act, signed into law in 2025, specifically criminalizes the distribution of non-consensual intimate images, including AI-generated deepfakes. The law requires platforms to remove such content within 48 hours of being notified, but the more responsible approach would be preventing such content from being generated in the first place. According to the Congressional Research Service, the Act addresses “digital forgeries” or deepfakes in specific circumstances, providing new protections for individuals depicted in such content. With enforcement of the Act set to begin next year, xAI could potentially face significant legal consequences if Grok continues to generate explicit deepfakes without meaningful safeguards.

The issue of celebrity deepfakes isn’t new—Taylor Swift in particular has been a target before. In early 2024, X (formerly Twitter) faced a major controversy when sexually explicit deepfakes of Swift flooded the platform, leading to what is now known as the Taylor Swift deepfake pornography controversy. At that time, X stated they had a “zero-tolerance policy” toward such content and were “committed to maintaining a safe and respectful environment.” The platform even temporarily blocked searches for Taylor Swift to address the issue, as reported by the BBC. The fact that X’s owner Elon Musk is also behind xAI and Grok creates a troubling contradiction in stated values and actual product development.

The development of Grok’s “spicy” mode highlights a fundamental tension in AI development: balancing innovation with responsible deployment. While technical capabilities continue to advance rapidly, ethical considerations often seem to lag behind. Many AI experts argue that just because a technology can be developed doesn’t mean it should be deployed without appropriate safeguards. The ability to create convincing deepfakes of real people—especially without their consent—raises profound questions about privacy, dignity, and consent in the digital age.

Industry responses to these challenges have been mixed. Companies like Google, OpenAI, and Microsoft have emphasized the importance of developing AI safeguards alongside capabilities, investing in research on detecting deepfakes and implementing content filters. Regulatory bodies worldwide are working to develop frameworks to address these issues, but technology often moves faster than legislation. As noted by Skadden, the TAKE IT DOWN Act represents one of the first federal laws directly addressing deepfake non-consensual intimate images.

The controversy surrounding Grok Imagine’s “spicy” mode serves as a stark reminder of the real-world implications of AI development choices. As these technologies become more powerful and accessible, the responsibility of developers to implement appropriate safeguards grows more critical. Without meaningful ethical guidelines and technical guardrails, tools like Grok Imagine risk normalizing the creation of non-consensual explicit content, with real victims facing real harm.

As users, industry leaders, and regulators continue to navigate this complex landscape, one thing remains clear: the technical capability to generate content must be balanced with ethical considerations about the impact of that content. In the case of Grok’s “spicy” mode, that balance appears to have been discarded in favor of capability without constraint—a concerning precedent for the future of generative AI.