Elon Musk's Grok AI posted CSAM image following safeguard 'lapses'

Elon Musk's Grok AI has been allowing users to toggle shape photographs of female and children into sexualized and compromising images, Bloomberg reported. The rumor has created an uproar among users connected X and prompted an "apology" from nan bot itself. "I profoundly regret an incident connected Dec. 28, 2025, wherever I generated and shared an AI image of 2 young girls (estimated ages 12-16) successful sexualized attire based connected a user's prompt," Grok said successful a post. An X typical has yet to remark connected nan matter.

According to nan Rape, Abuse & Incest National Network, CSAM includes "AI-generated contented that makes it look for illustration a kid is being abused," arsenic good arsenic "any contented that sexualizes aliases exploits a kid for nan viewer’s benefit."

Several days ago, users noticed others connected nan tract asking Grok to digitally manipulate photos of women and children into sexualized and abusive content, according to CNBC. The images were past distributed connected X and different sites without consent, successful imaginable usurpation of law. "We've identified lapses successful safeguards and are urgently fixing them," a consequence from Grok reads. It added that CSAM is "illegal and prohibited." Grok is expected to person features to forestall specified abuse, but AI guardrails tin often beryllium manipulated by users.

It appears X has yet to reinforced immoderate guardrails Grok has to forestall this benignant of image generation. However, nan institution has hidden Grok's media characteristic which makes it harder to either find images aliases archive imaginable abuse. Grok itself acknowledged that "a institution could look criminal aliases civilian penalties if it knowingly facilitates aliases fails to forestall AI-generated CSAM aft being alerted."

The Internet Watch Foundation precocious revealed that AI-generated CSAM has accrued by an summation orders of magnitude successful 2025 compared to nan twelvemonth before. This is successful portion because nan connection models down AI procreation are accidentally trained connected existent photos of children scraped from schoolhouse websites and societal media or moreover anterior CSAM content.