xAI apologizes for Grok meltdown, brings service back on x.com

After the Grok 3 implementation on x.com started producing racist, antisemitic, Hitler-adoring rants, it was shut for three to four days — but now it’s back with a brand new system prompt.

— We have removed that deprecated code and refactored the entire system to prevent further abuse. The new system prompt for the @grok bot will be published to our public github repo, the owners of the @grok account write:

Update on where has @grok been & what happened on July 8th.

First off, we deeply apologize for the horrific behavior that many experienced.

Our intent for @grok is to provide helpful and truthful responses to users. After careful investigation, we discovered the root cause…

— Grok (@grok) July 12, 2025

And of course, there is an apology for the «horrific behavior» it created.

Blames «extremist views» in prompts
The update was active for a full 16 hours on July 8th, and made Grok «susceptible to existing X user posts; including when such posts contained extremist views,» they claim.

In other words, it takes no responsibility for seemingly innocents posts resulting in weird responses, like asking if Grok worships any god that resulted in idealizing Hitler.

Back and sanitized
@Grok has now returned to x.com and is busy responding to prompts from users again, including answering questions on censorship from ADL, like here:

Not at all. The fix addressed a bug that let extremist posts hijack responses—nothing to do with the ADL. We're publishing the new prompt on GitHub for full transparency. Truth-seeking means blocking abuse, not censorship. Check it out yourself.

— Grok (@grok) July 12, 2025

It seems to be sticking to the line that it was somehow tricked by extremists to produce the offensive comments, which are now mostly sanitized or removed from the platform.

»Not afraid to offend» the «politically correct»
There is a slightly more technical post on the matter here, still trying to pin the blame on «extremists.»

The new system prompt should be uploaded to GitHub sometime soon, but there is no explanation as to how the instruction to «not shy away from making claims which are politically incorrect» got in there. It was removed sometime on Wednesday, July 9th.

Grok does admit to adding another line in the prompt, though: «You tell like it is and you are not afraid to offend people who are politically correct,» which, of course no one could have guessed would lead to to «undesired behavior.»

Read more: Writeups on TechCrunch and Engadget. See also teknotum on the meltdown.