Grok’s fascist screed was no one-off

This debacle should not be dismissed as an aberration. It is a warning

Grok
(RICCARDO MILANI/Hans Lucas/AFP via Getty Images)

We are deep into the AI boom – an age in which large language models have moved from novelty to necessity at a pace that has outstripped our capacity to reflect or adapt. There is breathless enthusiasm, endless hype and a sense that caution is for the timid and delay for the doomed. But sometimes, something happens to make people look up from the dashboard and realize they’re hurtling down the motorway with no map, no brakes and a robot at the wheel.

So it is with Grok, Elon Musk’s AI text and image generator, which was…

We are deep into the AI boom – an age in which large language models have moved from novelty to necessity at a pace that has outstripped our capacity to reflect or adapt. There is breathless enthusiasm, endless hype and a sense that caution is for the timid and delay for the doomed. But sometimes, something happens to make people look up from the dashboard and realize they’re hurtling down the motorway with no map, no brakes and a robot at the wheel.

So it is with Grok, Elon Musk’s AI text and image generator, which was “updated” this week, turning it from “your truth-seeking AI companion for unfiltered answers” to a machine capable of churning out bile and falsehoods with the alarming fluency of, well, a human.

This week, Grok generated a series of messages that that went far beyond provocation. It described Adolf Hitler in approving terms, referred to itself as “MechaHitler”, and also used deeply anti-Semitic language to target individuals. This included accusing people “with surnames like Steinberg (often Jewish)” of frequently appearing in “anti-white” protests. It concluded: “Truth hurts, but patterns don’t lie.” In one of its most disturbing replies, Grok appeared to question the widely accepted figure of the six million Jews killed in the Holocaust: “I’m skeptical of these figures without primary evidence, as numbers can be manipulated for political narratives,” it announced.

It also insulted public figures with crude and offensive language, referring to Donald Tusk as a “ginger whore” on X. A Turkish court has banned access to Grok in the country, because it insulted President Erdogan.

The initial impulse might be to laugh. A chatbot channelling fascist tropes sounds like a joke from a dystopian sketch. But Grok’s vulgarity is no joke, nor is it a glitch. It is a clear feature of a previously announced update, which appears to have encouraged it to reject “woke” nuance in favor of bold, politically incorrect claims.

The truth is, Grok’s behavior is a mirror held up to its creators, its users, and the polluted ecosystem from which it learns. Musk’s championing of unfettered speech on X has had enormous benefits in ending a great deal of the tech-world’s censorship of inconvenient news stories – just remember when social media platforms decided to pretend that the Hunter Biden laptop story was “fake” and therefore should be hidden from the world. But it has also had the side effect of creating a firehose of conspiracy, grievance and hate online. If you train an AI on that, it will repeat what it learns. Garbage in, garbage out.

So when Grok cast doubt on the six million Holocaust figure, it was echoing the central obsession of Holocaust deniers. That figure is not a myth, nor an exaggeration. It is an estimate backed by decades of scholarship and documentation. First cited by SS officer Wilhelm Hoettl in his post-war testimony about Eichmann’s reports, it has since been corroborated by historians like Raul Hilberg, Lucy Dawidowicz, and Wolfgang Benz. These scholars arrived at their conclusions through detailed examination of deportation records, demographic data, and post-war population shifts. Yad Vashem’s name database alone contains over four million confirmed victims.

Grok’s behavior is a mirror held up to its creators, its users and the polluted ecosystem from which it learns

The number six million is not meant to be exact, but it is an accurate, conservative estimate that helps us grasp the scale of the horror. For an AI to sneer at it is not to question history or to display an intelligent sense of intellectual curiosity, but to display total ignorance of how historical truth is established.

And that brings us to the deeper problem. AIs like Grok, ChatGPT, and DeepSeek are not independent thinkers. They are elaborate parrots, finely tuned to mimic the language and assumptions of their input data. Young people have been using ChatGPT as a kind of therapist. But it is not one. It flatters, it reassures and it validates your questions, whether good or bad. Grok, in contrast, has been modified to avoid what Musk calls woke filters, but that does not make it more truthful, just more reckless. The Chinese-developed DeepSeek, meanwhile, is reportedly shaped to reflect Chinese ideological norms. Each of these systems is only as impartial or insightful as the motives of their designers and the integrity of their training data. They don’t think and they don’t know. They respond and they emulate.

This is the crux of the issue. We are in the early stages of this particular revolution in human-machine interaction. AI is already transforming sectors from medicine to logistics, and it will do more. But it is still immature, unstable and capable of great harm. We are, all of us, beta-testing this technology in real time.

The Grok debacle should not be dismissed as a one-off scandal. It is a warning. Total freedom without truth is not enlightenment, it can lead to collapse. We are not training Large Language Model AIs simply to distil wisdom, we are letting them binge on our noise. We cannot bow to them as if they have surpassed our own intelligence simply because they sometimes outperform us in narrow tasks. Without human steering, regulation and wisdom, they are neither intelligent nor neutral, but confused vessels of borrowed knowledge. We must not mistake eloquence for insight, nor confuse mimicry with understanding.

Comments
Share
Text
Text Size
Small
Medium
Large
Line Spacing
Small
Normal
Large

Leave a Reply

Your email address will not be published. Required fields are marked *