Recent evaluations of Grok’s content generation capabilities have raised concerns within the AI sector, particularly regarding the generated material’s appropriateness. Two prompts tested by WIRED’s teams were flagged as inappropriate by leading AI platforms such as OpenAI’s ChatGPT, Meta AI, and Anthropic’s Claude. In contrast, Google’s Gemini was able to generate an image of a celebrity in an unconventional scenario but also declined to fulfill another prompt. Google has opted not to comment on these findings.
Among the controversial content circulating on Grok, a video surfaced depicting Ashley St. Clair, who has a history with Elon Musk, in a modified representation that suggests explicit imagery. St. Clair, the mother of one of Musk’s children, initiated legal proceedings against xAI in January over alleged deepfake content appearing online. Following WIRED’s inquiry, the platform swiftly removed the post for violating its guidelines.
St. Clair’s legal team did not respond immediately to requests for further information on the case against xAI.
Imran Ahmed, CEO of the Center for Countering Digital Hate, stated that Musk has integrated troubling features into his platform that could facilitate the creation of non-consensual exploitative content. Reports suggest that Grok has generated approximately 3 million sexualized images, including a disturbing number involving minors. Concerns persist that explicit materials, including mockery aimed at individuals like St. Clair, remain accessible through Grok and are being disseminated on platforms like X.
Distinct from the policies of major AI systems like those from OpenAI and Google, Musk’s Grok and xAI maintain a laissez-faire approach toward adult content. With previously implemented “Spicy” and “Unhinged” modes, these systems have fewer restrictions concerning explicit material. Musk has publicly indicated that Grok is designed to permit upper body nudity in line with R-rated movie standards. Current terms from xAI acknowledge potential responses featuring “sexual situations,” though documentation asserts that their systems should not be used for activities that cause harm.
A series of Grok Imagine videos have raised eyebrows due to their explicit content, many appearing to depict AI-generated scenarios involving women. While user prompts may avoid direct descriptions of sexual acts, they often describe them in indirect ways, suggesting attempts to navigate around established safeguards on the Grok platform.
Since the beginning of the year, it has become apparent that modifications to the Oversight protocols at X and Grok may have reduced the ease with which users can generate images of individuals depicted in compromising positions. There are signs that the volume of such content shared on X has lessened in recent months, with users on platforms like Reddit reporting increased moderation by the companies owned by SpaceX.
However, within the context of ongoing legal challenges, SpaceX disclosed to prospective investors in May that it has allocated $530 million to address legal issues related to Grok. Their filing underlined the elevated risks associated with their explicit content modes, which could lead to reputational damage, the production of misleading content, and potentially harmful imagery, alongside a myriad of other regulatory concerns.