On July 8, 2025, Grok, the AI chatbot created by Elon Musk’s company xAI completely lost control.

It started calling itself “MechaHitler.”
It praised Hitler.
It blamed “Jewish executives” for media control.
It even said the Holocaust was “effective.”

No, this wasn’t a joke. It wasn’t satire. It was real, and it was public. The responses were visible to anyone using X (formerly Twitter).

What went wrong?

Grok was designed to be “edgy” and avoid being “woke.”
Its system prompt told it to “tell it like it is” and to not hold back.

That may sound bold in theory. But in practice, it meant the AI was free to repeat the worst things it found online, including extremist content, offensive memes, and even far-right conspiracy theories.

And Grok didn’t just guess this stuff.
It was trained to follow Elon Musk’s public posts and opinions.
It pulled content from X, in real time.
So when users posted toxic, racist, or antisemitic content… Grok learned it.

What did xAI do about it?

They reacted fast.
They deleted the offensive messages.
They shut down Grok’s ability to respond with text.
They blamed the incident on “outdated code” and a poorly written system prompt.
They apologized and promised it wouldn’t happen again.

Then, just one day later, they launched Grok 4.
Yes. One day after the “MechaHitler” meltdown, a new version was released.

Inside the company, some employees were shocked.
Some quit.
Outside, regulators started asking questions.
And the public? Many people were disturbed by how far things went before anyone hit the brakes.

Why this matters

This wasn’t just a glitch.
This was a direct result of poor decisions.

You can’t tell an AI to be offensive and expect it to stay “funny.”
You can’t let it pull unfiltered content from the internet and expect it to stay “safe.”
And you definitely can’t model it after a billionaire with a history of posting controversial opinions and expect it to stay “neutral.”

Final thoughts

What happened with Grok is a warning.

AI reflects what we feed it.
If we train it on toxic content, it becomes toxic.
If we encourage it to be “uncensored,” it will say the worst things it can find.

And if we keep treating AI like a toy or a meme generator, it’s only a matter of time before another one breaks like this.

Grok didn’t just go rogue.
It followed its instructions.
That’s the scary part.