Human
Did Grok finally go too far?
Grok did it again.
We've already seen xAI's... idiosyncratic LLM behave erratically in the past, but now to Hitler Grok, antisemitic Grok, white genocide Grok and pro-Kremlin Grok we can add child-porn Grok and creepy stalker Grok.
This latest controversy, however, feels different. In previous iterations of the never-ending conflict between Elon Musk and the woke mind virus, the erstwhile prince of Mars has tended to double down on defence of his frequently out of control AI assistant because (gestures vaguely) free speech. This time, however, after it was reported that Grok was allowing users to tag images of women and children to be undressed without consent, xAI quickly restricted its image-generation abilities to premium subscribers. After all, when your billionaire owner devoted a huge chunk of 2025 to haranguing the UK government over its failures to investigate predatory grooming gangs, suddenly becoming the conduit for online predatory grooming gangs is not a great look.
What is most remarkable about all of this is just how unremarkably predictable it all was. One of the biggest complaints about Grok 1 by the terminally online right when it was first launched in 2023 was that it tended to parrot the progressive opinions of publications such as the New York Times. Since then, as in everything, Musk has taken a particular, perverse pride in making Grok a mirror image of his contrarian social media platform, with any opposition treated as pearl-clutching opponents of the principles of the First Amendment. That he may occasionally be right does not detract from the fact that he is usually very, very wrong.
The sheer weirdness of the world in which he and his minions live was epitomised in the willingness of many of them to pay $200 per month for the privilege of using Grok Ani, animated companions including a distinctly underdressed anime girl who was "super crazy in love" with (paying) users and whose age was specified as 22, but who indulged a gothic lolita fashion that made her appear much younger. About the same time, Grok Imagine launched a "spicy mode" which pooh-poohed the milquetoast concerns of competitors such as OpenAI and Google. When image generation capabilities were rolled out to the fine folk who comprise the core user base of X, it was only a matter of time before they would turn them to nefarious purposes.
Guardrails? How do you spell that?
Yann LeCun, Meta's chief AI scientist, once remarked that the task of computer scientists was to "hardwire the architecture of AI systems so the only actions they can take are towards completing objectives we give them, subject to guardrails." There is a certain irony that Meta, of all companies, should have made observations on the ethics of computing, but the point is one that - every once in a while - all the major AI companies at least pay lip service to in their grand pursuit of a god in the box.
While Musk has supported regulation of AI at least as recently as mid 2024 when he welcomed the proposed California saftegy bill (SB 1047), more often he has been deeply ambivalent. In comments reported by Axios around the same time, he remarked that he considered guardrails to be a form of censorship and made the gnomic comment that AI "should not be taught to lie", and that attempts to align it with human social values were - you guessed it - part of the woke mind virus.
Making AI alignment a partisan issue seems to have been very much part of the radical transformation that Musk underwent during and following his purchase of Twitter in 2023, but he may also be starting to realise that his increasingly libertarian attitude is revealing itself as its own Faustian pact. Although he has inveighed against those using Grok to make illegal content and said that they will be punished, the sheer wilfulness of X users (and, indeed, Grok itself) stems from his attempts to shape public opinion - human and artificial - in his own image.
And the headlines circulating this week indicate just how precarious this route is: whether it is "Grok makes sexual images of kids as users test AI guardrails" or "Dark web users cite Grok as tool for making 'criminal imagery' of kids", the storm surrounding xAI means that Musk may have handed UK and EU governments their perfect opportunity for revenge. Not too many people were worried a month ago when Grok (just handed control over autonomous driving for Tesla vehicles) said it would run over a billion children to avoid hitting Elon Musk. That could be passed off as a joke, but in the age of the Epstein files having your pride and joy make images of sexualised children is unforgiveable, whether your politics are progressive or conservative.