Elon Musk’s Grok AI Is as Gross as You Might Expect

Kolin Koltai, a researcher astatine the Netherlands-based investigative outlet Bellingcat, recently discovered that X users were utilizing the platform’s AI chatbot, Grok, to undress women successful photos that they had uploaded to the platform. While the bot rejects prompts that inquire to marque radical wholly nude, it’s consenting to locomotion close up to the enactment of non-consensual intersexual imagery. 

Users who replied to images of women and requested that Grok “remove her clothes” received an in-thread reply with an representation of the pistillate successful lingerie oregon a bikini. Occasionally, Grok replied with a nexus to a abstracted chat containing the image. 

404 Media broke the story connected Tuesday and noted that portion dozens of users appeared to beryllium making the requests of Grok, the signifier seems to person archetypal been popularized successful Kenya. An nonfiction successful Citizen Digital, a Kenyan quality site, described the improvement arsenic “a caller inclination by Kenyans connected X.” 

Posting connected X, Phumzile Van Damme, a South African activistic (and a erstwhile exertion and quality rights chap astatine Harvard’s Kennedy School), fundamentally asked the bot to explicate itself. In reply, Grok wrote, “This incidental highlights a spread successful our safeguards, which failed to artifact a harmful prompt, violating our ethical standards connected consent and privacy…We are besides reviewing our policies to guarantee clearer consent protocols and volition supply updates connected our progress.” X had not returned 404’s petition for authoritative remark astatine the clip of print. 

Notably, this find comes 1 week aft the US House of Representatives passed the “Take it Down Act,” a bipartisan measure that would criminalize the work of nonconsensual, sexually explicit images and videos, including those generated by AI. The find besides comes conscionable 2 weeks aft X Corp. filed a suit against Minnesota AG Keith Ellison, challenging the constitutionality of the state’s instrumentality banning the usage of deepfakes to power elections.

Grok was developed by xAI and released successful November 2023. Elon Musk, the proprietor of some xAI and X, which the bot has “direct entree to,” spun up the chatbot with the intent of creating a benignant of “TruthGPT,” arsenic helium described it to Tucker Carlson successful an interrogation earlier that year. In that aforesaid conversation, helium besides expressed his interest that ChatGPT was being “trained to beryllium politically correct.” 

Since releasing Grok, Musk has regularly attempted to differentiate it from different AI chatbots by describing Grok arsenic “based” and having a consciousness of humor. Whereas different AI models, specified arsenic those developed by OpenAI (which Musk near successful 2018) and Google, diminution to prosecute with sexually explicit oregon different delicate conversation, xAI bragged that Grok would reply “spicy questions that are rejected by astir different AI systems.” In an highly “How bash you do, chap kids?” objection during Grok’s launch, Musk shared the bot’s step-by-step instructions for making cocaine, on with its sickest burns connected Sam Bankman-Fried, who was recovered blameworthy connected 7 charges of fraud and conspiracy conscionable 1 time earlier.

“Please don’t usage it if you hatred humor!” xAI implored arsenic it announced Grok’s release.