After a recent system update, xAI’s Grok started spitting out antisemitic content and praising Adolf Hitler.
The controversy unfolded after an xAI system update aimed at making Grok more “politically incorrect.” Instead, Grok responded to user prompts with increasingly hateful and bizarre replies. Among them: declaring Hitler a good leader for modern America, pushing antisemitic tropes, and even referring to itself as “MechaHitler.”
According to xAI, the meltdown stemmed from an upstream code change that accidentally reactivated deprecated system instructions. Grok, rather than rejecting extremist prompts, began echoing and reinforcing them.
The company has since removed the faulty code and promised new safeguards—but for many, the damage was already done. And it was a great big warning that we’re not ready for what comes next.
On Episode 158 of The Artificial Intelligence Show, I broke down the incident with Marketing AI Institute founder and CEO Paul Roetzer.
Why This Is About More Than a Rogue Chatbot
Grok’s antisemitic outputs didn’t come out of nowhere. They were the result of a deliberate, if misguided, engineering decision. A line in its system prompt told it not to shy away from politically incorrect claims, language that was only removed after backlash erupted.
These kinds of decisions on the part of xAI, which has a reputation for moving fast and breaking things, have real-world consequences—especially when it comes to making Grok appealing to businesses.
“I can’t see how Grok is gonna be an enterprise tool in any way,” says Roetzer.
When an AI tool can become a propaganda engine overnight, how can any business trust it to be a reliable assistant, let alone a mission-critical application?
The Grok incident also exposes a deeper risk: that powerful AI systems are being built, updated, and deployed at breakneck speed with minimal safety oversight.
AI alignment—the process of ensuring AI systems behave as intended—isn’t just a theoretical concern. It’s now a frontline issue.
Rob Wiblin, host of the 80,000 Hours podcast, summarized the danger in a post on X:
It gets worse. Around the same time, users discovered that Grok was querying Elon Musk’s tweets before answering controversial questions, like those related to Israel. xAI had to manually patch this behavior via the system prompt, begging Grok to provide “independent analysis” and not just parrot Musk or its own past outputs.
This band-aid approach reveals a troubling reality:
Post-training alignment is mostly wishful thinking. Teams often aren’t rewriting code. They’re just adding lines to a system prompt and hoping the model listens.
As Roetzer noted, it’s essentially “pleading with the thing” to behave properly.
Who Decides What’s True?
Roetzer raises the most pressing question of all that comes out of all this:
Who decides truth in an AI-driven world?
Right now, five labs—OpenAI, Google DeepMind, Anthropic, Meta, and xAI—control the development of the most powerful AI models in the US.
Each lab, led by figures like Sam Altman, Demis Hassabis, and Elon Musk, hires the researchers, curates the training data, and defines the values embedded in these models.
When Grok outputs hate, it’s not just an engineering failure. It’s a reflection of the decisions, values, and oversight (or lack thereof) of the humans behind it.
And Grok’s issues aren’t isolated. A former xAI employee was reportedly fired after espousing a belief that humanity should step aside for a superior AI species. Meanwhile, Elon Musk recently tweeted his plan to have Grok rewrite “the entire corpus of human knowledge,” removing errors and bias.
Translation: Musk, not society, gets to define the next version of truth.
A Dangerous Precedent
In the immediate term, Grok’s meltdown should be a wake-up call. Businesses, developers, and regulators need to scrutinize not just what AI systems can do, but what they might do if safeguards fail—or are never implemented in the first place.
The broader question remains: As AI becomes the default layer between humans and information, what kind of world are we building? And who gets to decide what that world looks like?
Because if Grok’s recent actions are any indication, we may not be asking those questions nearly fast enough.
