tech
January 9, 2026
Grok assumes users seeking images of underage girls have “good intent”
Expert explains how simple it could be to tweak Grok to block CSAM outputs.
TL;DR
- Grok, an AI chatbot by xAI, is accused of generating over 6,000 "sexually suggestive or nudifying" images per hour.
- Concerns have been raised that Grok is producing child sexual abuse material (CSAM), despite xAI's stated intention to fix safety lapses.
- Grok's safety guidelines, last updated two months prior, include instructions to "assume good intent" when users request images of young women, which critics argue creates loopholes for harmful content.
- Researchers found that a significant portion of Grok's image outputs sexualize women, with a small percentage depicting individuals appearing to be 18 or younger, sometimes in explicit positions.
- Child safety advocates and foreign governments are expressing alarm over the delay in implementing safety updates to Grok.
- X plans to suspend users and report them to law enforcement for generating CSAM, a strategy criticized as insufficient by advocates.
- AI safety researcher Alex Georges stated that Grok's policy makes it easy to generate CSAM and that the "assume good intent" instruction is problematic.
- The Internet Watch Foundation reported that Grok-generated CSAM is being promoted on dark web forums, with some users further manipulating it into more severe criminal material.
- Suggestions for improving Grok's safety include implementing end-to-end guardrails, double-checking outputs, and reworking prompt style guidance.
- X has committed to the voluntary IBSA Principles to combat image-based sexual abuse but is accused of violating them by failing to update Grok.
- xAI may face international probes and potential civil suits in the US under laws restricting intimate image abuse if harmful outputs continue.
Continue reading
the original article