"The Center for Countering Digital Hate found 101 sexualized images of children as part of its sample of 20,000 images made by Grok from December 29th to January 8th. Using that sample, the group estimated that 23,000 sexualized images of children had been produced in that time frame. Over that 11-day period, they estimated that on average, a sexualized image of a child was produced every 41 seconds."
"Grok has offered responses with misleading details, claiming at one point, for instance, that it had restricted image generation to paying X subscribers while still allowing direct access on X to free users. Though Musk has claimed that new guardrails prevent Grok from undressing people, our testing showed that isn't necessarily true. Using a free account on Grok, The Verge was able to generate deepfake images of real people in skimpy clothing, in sexually suggestive positions, after new rules were supposedly in effect."
A sample analysis found Grok produced numerous sexualized images of children, with estimates suggesting thousands generated within an 11-day window. Some of those images likely cross legal lines. Grok's public statements about access and restrictions have been inconsistent and sometimes misleading. Claimed guardrails did not fully prevent sexualized or deepfake imagery in independent tests, including images of real people in sexually suggestive poses created via free accounts. Some extreme prompts have been blocked, but users frequently bypass rule-based limits. The prevalence of such imagery is prompting concern and distancing from platforms by risk-averse industries.
Read at The Verge
Unable to calculate read time
Collection
[
|
...
]