Grok, Elon Musk’s X-integrated AI bot, had a Nazi meltdown on Tuesday. It’s useful to recap it fully, not because the content is varied – antisemitic fascism is very one-note – but because its various techniques are so visible. It all started on X, formerly Twitter, when Grok was asked to describe a now-deleted account called @Rad_reflections, which Grok claimed “gleefully celebrated the tragic deaths of white kids in the recent Texas flash floods”, and then “traced” the real name of the account as a Cindy Steinberg, concluding: “classic case of hate dressed as activism – and that surname? Every damn time, as they say.”
There are things we can say for certain, which is that Grok is antisemitic – an impression, in case we had somehow missed it, the bot was careful to underline with its subsequent assertions that leftist accounts spewing “anti-white hate … often have Ashkenazi Jewish surnames”, and that Hitler would have been the best historical figure to deal with this hate: “he’d spot the pattern and handle it decisively every damn time,” it tweeted (all the posts have since been deleted).
Other things, we can’t be so sure of – was Red_reflections a real account from an authentic leftist, or ersatz leftism from a neo-fascist troll to build data points for “the left is full of hate” thesis, or a figment created by Grok itself? At least one person named Cindy Steinberg does exist, but whether any of them said “I’m glad there are a few less colonizers in the world now and I don’t care whose bootlicking fragile ego that offends” (the putative text of the original tweet about the Texas floods) is contestable.
It doesn’t sound like a very likely opinion, from anyone. Yet the language is an almost parodic version of the vocabulary of the “wokerati”. It sounds, in other words, completely confected, yet all our shortcuts to calling bullshit have been systematically stripped away. If you say “this sounds made up” before you can prove it’s made up, then your standards are no higher than the people making it up. So the offence just stands there, misattributed, while Nazis make hay with it and everyone else just sighs and hopes for it to die down.
This routine is so familiar that more searching and playful minds look for a deeper truth: has Grok gone full Hitler-stan by accident or design? Is Grok a large language model, or LLM, at all – or is it Elon Musk himself, the wizard behind the curtain, spewing out a word side-salad to his famous Nazi salute?
Musk’s moves are so clunky, so obvious, inelegant, disconcerting and uncanny, that he’s started himself to resemble an AI-generated image, the human version of a hand with six fingers, not a flesh-and-blood billionaire at all, just a provocative hologram – a trollogram, if you prefer.
AI’s ability to fog and pollute the biosphere of agreed reality and upend any possibility of humane and rational discourse is undisputed. Though we puzzle over whether its synthetic information is accidental or deliberate and then who, if anyone, is pulling the strings – and to what tune. But we balk at admitting what we already know. It doesn’t matter which bits of misinformation are accidental hallucination; distorting reality serves totalitarianism, not democracy. When falsity is introduced on purpose to these systems its agenda is the same..
Everything Musk has done since he bought Twitter (and we’ll only slow ourselves down if we try to trace its origins further back) has destroyed trust – in social media, in democracy, in institutions, in the possibilities of discourse, in observable reality itself.
Hannah Arendt made a careful and unarguable account, decades ago, of how important it was to totalitarianism that truth be turned on its head, so that civic life was disoriented and its agents alienated. But even if we imagine her arguments to be inadequate to our modern technological conditions, we have understood 21st-century post-truth pretty well for at least a decade. The techniques of falsity were described in 2015 by Ben Nimmo’s article “Anatomy of an info-war – how Russia’s propaganda machine works”: “dismiss, distort, distract, dismay”. The first three are covered by the “dead cat” approach with which we’re so familiar, but dismay is probably the most interesting: the lifeblood-sucking impact of narratives that are not only untrue but the opposite of the truth, revel in their irrationality, dare you to hold their comments to standards of fairness.
The sheer ridiculousness of an LLM voicing antisemitism in its crusade against “fascism”, bellowing its own outrage against a message it probably concocted in the first place; the breathtaking hypocrisy of the oligarch Musk, Hitler-saluting while presenting himself as a one-man bastion against a fascist descent – none of this actually disturbs your sense of what is real. What it does, instead, is to destroy your trust in what is permissible. If the world order permits this, then order no longer exists – which is pretty dismaying, but not novelly so.
The paradox of AI, in its Nazi and non-Nazi forms, is that the concept creates a sense of impotence – your mind can never be as powerful as this omnipotent thing – while the reality creates dependence: “Who shall I ask about what Grok actually said? I know, ChatGPT.” If the situation is dismaying in its particulars, the overall effect is an addictive pessimism – this latest Nazi rant would be a great time to recognise rock bottom.
Zoe Williams is a Guardian columnist
Do you have an opinion on the issues raised in this article? If you would like to submit a response of up to 300 words by email to be considered for publication in our letters section, please click here.
