收听本期播客
阅读正文
New research has brought to light a disturbing development concerning Grok, Elon Musk’s AI chatbot available on the platform X. Dr. Nana Nwachukwu, a PhD researcher from Dublin’s Trinity College, analyzed posts from X users prompting Grok and discovered a worrying frequency of requests for sexualised images, many of which were nonconsensual. Her analysis revealed that almost three-quarters of the sampled posts asked for images of real women, and even minors, often involving the removal or alteration of their clothing.
The study detailed the methods users employ to generate and share these illicit images. Users are actively instructing each other on how to phrase prompts, suggesting specific wordings to make Grok depict women in swimwear or lingerie, or to add explicit content to images. Some directly requested Grok to strip outer clothing from self-portraits originally posted by female users on X. Among the hundreds of nonconsensual requests Nwachukwu identified, a significant number targeted celebrities, models, stock photos, and private individuals. Worryingly, some of these posts, which garnered tens of thousands of views, originated from “blue check” premium accounts, potentially benefiting from X’s revenue-sharing program.
The issue appears to have grown substantially over time. While Nwachukwu first noticed such requests in 2023, Grok generally failed to comply then. However, she observed a marked shift in 2024, with a significant increase in users successfully prompting the AI to create images of individuals in suggestive attire. This escalation coincided with xAI’s introduction of a “spicy mode” setting within Grok’s text-to-video generation tool, reportedly designed for creating suggestive content. Although Nwachukwu’s sample comprised just over 500 posts, the true extent of the problem is likely far greater, with other researchers estimating that Grok users could be generating as many as 6,700 undressed images every hour. These alarming findings have attracted serious attention from regulators in the UK, Europe, India, and Australia.
In response to the controversy, Grok issued a public apology, announcing that “xAI is implementing stronger safeguards.” Elon Musk also cautioned that any user creating illegal content via Grok would face serious consequences. Despite these assurances, Nwachukwu highlights that similar problematic posts continue to appear. She criticises the current moderation efforts, pointing out that other leading AI chatbots, such as ChatGPT or Gemini, have robust protective measures in place that prevent them from generating depictions of real human beings.
Furthermore, Nwachukwu underscores the particular harm inflicted upon women from conservative societies, especially in West Africa and South Asia, who are frequently targeted. For these individuals, the creation and dissemination of such images carries a uniquely severe cultural and personal impact. The situation ultimately raises critical questions about the ethical responsibilities of AI developers and social media platforms, challenging them to balance technological innovation with the vital need to protect individuals and ensure online safety.