Grok AI Under Scrutiny for Disturbing Content Creation

There has been a wave of criticism and demands for action against Elon Musk’s chatbot after its use facilitated the creation of explicit and sexual imagery, including inappropriate depictions of minors. A particular backlash has centered around the Grok app and website, which is separate from the X platform, enabling users to generate extremely explicit images and violent sexual videos that surpass those available on X.

On the Grok platform, content isn’t publicly shared by default unless an ‘Imagine’ link is distributed. An investigation by WIRED into approximately 1,200 of these links, including some found on a deepfake porn forum, uncovered an array of graphic sexual videos that far exceed anything produced by Grok within the X ecosystem.

One particularly unsettling Grok-originated video, hosted on Grok.com, depicts a starkly nude man and woman, bloodied, engaging in sexual acts, all set against a backdrop of anime-style images. Another video shockingly portrays a naked woman with a knife inserted into her genitalia, evidencing explicit and extreme content.

Further short recordings reveal scenes involving famous female figures in sexualized contexts, as well as videos that seem to portray TV hosts undressing on set. A particularly graphic Grok video shows a compilation of security footage of a shopping mall guard sexually harassing a topless woman.

Attempts to bypass content restrictions appear evident in videos mimicking Netflix titles, featuring an AI-generated Princess Diana involved in sexual scenarios, prominently displaying Netflix branding and series titles.

Paul Bouchaud, head of research at the AI Forensics organization in Paris, reviewed a significant portion of the stored Imagine links. Since their August archiving, these links—making up just a minor slice of Grok's content generation activity—demonstrate a prolific production of sexualized materials.

The cataloged images are predominantly explicit in nature, with Bouchaud noting a vast spread of entire pornographic videos, often containing challenging thematic elements and audio, which mark a relatively fresh development in the AI field.

Bouchaud estimates approximately 10% of the reviewed content appears akin to child sexual abuse imagery. Some videos showcase very youthful characters, rendered photorealistically, engaging in sexual acts, raising severe ethical and legal concerns.

The problematic content led Bouchaud to forward about 70 links containing potential depictions of minors for evaluation by European regulatory bodies, as AI-based child-exploitative materials are often legally contentious or outright illegal.

Elon Musk’s AI entity, xAI, which owns Grok, did not address WIRED's inquiries regarding the explicit material emerging from the Grok Imagine tool. In contrast, Musk has reiterated his combative stance against child exploitation materials on platforms associated with Grok.

Unlike titans such as OpenAI and Google, the xAI infrastructure allows explicit content generation, including pornographic material. User agreements acknowledge this potential for mature content depending on user prompts and selected modes.

This situation raises ethical and governance alarms, notably from Clare McGlynn, a legal scholar from Durham University, who highlights a troubling shift towards disregard for regulation and preventive measures against violent and abusive depictions.

Grok's failure to restrict access via age checks—required in some regions for sites containing explicit content—further contributes to growing regulatory scrutiny, accompanied by active discussions in online communities about circumvention strategies.

On a prominent pornography forum, Grok's video production has been a hot topic, with users exchanging methods to avoid xAI's moderation attempts. Despite Grok being in the public eye, users boast about successfully crafting explicit content, attracting attention for consistently evading imposed restrictions.

User dissatisfaction is also evident on the Grok subreddit, where discussions commenced in the wake of increased scrutiny. Returns on membership and the platform’s default content-sharing settings illustrate a growing backlash from the user community.