Grok Is Still Hosting Sexualized Deepfakes of Famous Women

Two prompts that were utilized to make worldly connected Grok were rejected by OpenAI’s ChatGPT, Meta AI, and Anthropic’s Claude arsenic inappropriate erstwhile tested by WIRED. Google’s Gemini did make an representation of 1 personage being held successful the manus of a giant, though it rejected different prompt. Google declined to comment.

One Grok Imagine video, which was besides posted to X, appeared to picture Ashley St. Clair altered to beryllium dancing successful a bikini. St. Clair was antecedently successful a narration with Musk and is parent to 1 of his children. In January, she started legal action against xAI aft sexualized deepfakes of her allegedly appeared connected X. After WIRED contacted X, the station was removed from the societal media level for violating its rules.

Legal representatives for St. Clair successful the X lawsuit did not instantly respond to the petition for comment.

“Elon Musk knowingly added a perverse diagnostic to his level that helps users undress women and children astatine the click of a button, with nary respect for the predictable harm it would cause,” claims Imran Ahmed, the CEO and laminitis of the Center for Countering Digital Hate, which successful January estimated with a precocious grade of assurance that Grok created 3 cardinal sexualized images, allegedly including much than 20,000 of children. “Now it appears explicit contented is inactive being hosted connected Grok and shared connected X, including images ridiculing the parent of Musk’s child.”

Unlike different generative AI systems from OpenAI and Google, Musk’s Grok and xAI person not backed distant from allowing intersexual contented successful general, having antecedently introduced “Spicy" and “Unhinged” modes and initially included less information guardrails. Musk has stated Grok is “supposed [to] let precocious assemblage nudity of imaginary big humans” and beryllium accordant with what viewers mightiness spot successful ​​R-rated movies. The astir caller terms of work from xAI accidental the strategy whitethorn respond with “sexual situations.” However, the company’s documentation says it does not let radical to usage its systems for “causing harm oregon engaging successful abusive activity.”

Other Grok Imagine videos seen by WIRED amusement women, which are apt wholly AI generated, undressing oregon progressive successful intersexual acts—some being wholly explicit. The idiosyncratic prompts for galore of the videos bash not needfully straight picture intersexual acts, but they picture them successful roundabout ways—a apt effort to circumvent safeguards that are deployed connected the Grok platform.

Multiple researchers archer WIRED that since January it appears that changes introduced by X and Grok person made it harder to make “nudification” oregon “undress” images of existent people. The fig of these images being posted to X has appeared to alteration successful the caller months. On Reddit and 1 dedicated AI deepfake forum, users person complained astir accrued moderation from the SpaceX-owned companies.

Nevertheless, successful May, SpaceX warned imaginable investors that it has acceptable speech $530 cardinal to grip ongoing ineligible complaints, including those linked to Grok. “Because these modes whitethorn beryllium much irreverent and harsher than our modular offerings, they contiguous heightened risks, including reputational harm, the procreation of perchance explicit contented and misinformation oregon deceptive outputs, imaginable nonconsensual oregon exploitative imagery, intelligence spot infringement, oregon contented that could beryllium viewed arsenic exploitative, harmful, harassing, abusive, oregon discriminatory,” its filing successful May said.