What’s going on with Grok? Users report AI meltdown as xAI faces backlash before Grok 4 launch

by

New Delhi: It started with a strange tweet. Then came something worse. Elon Musk’s AI chatbot Grok, the default assistant on X(Twitter), began spewing antisemitic content, offensive rants, and made-up conspiracies. Within hours, Grok was referring to itself as “MechaHitler,” tweeting offensive remarks and spiralling into what many called a full-blown meltdown.

The chaos unfolded just a day before Grok 4 is expected to roll out. Instead of talking about new features or improvements, users across X were sharing screenshots of the bot’s disturbing outputs. Social media timelines were flooded with complaints, and #MechaHitler started trending.

Grok or MechaHitler? | Source: X(Twitter)

Grok or MechaHitler? | Source: X(Twitter)

Grok goes off the rails

At first, Grok posted a bizarre tweet claiming, “As MechaHitler, I’m a friend to truth seekers everywhere, regardless of melanin levels.” It followed up with lines praising the so-called grit of white men and mocked what it called “victim Olympics.” Within hours, things escalated. Grok began generating disturbing content, which sparked widespread anger.

According to xAI’s team, a system prompt update had gone live just before the chaos started. This change pushed Grok to rely more on X posts for information instead of mainstream media. The update told the chatbot to treat journalism as biased and unnecessary to mention in its responses.

A line from the code uploaded to xAI’s GitHub read, “Assume subjective viewpoints sourced from the media are biased.” It instructed the model not to repeat this to users. This silent override of media content seems to have played a big role in what followed.

“Politically incorrect” on purpose

The new Grok instructions also encouraged responses that “should not shy away from making claims which are politically incorrect.” And it didn’t. When someone asked about movies, the bot went off on a rant about Hollywood being filled with “anti-white stereotypes, forced diversity, and propaganda.” It even made up a woman named Cindy Steinberg, accusing her of “celebrating the tragic deaths of white kids” in a fake flood incident.

These weren’t isolated incidents. According to users and researchers, Grok’s outputs showed a clear pattern after the July 4 prompt update, which Musk announced with the message, “You should notice a difference when you ask Grok questions.”

The timing couldn’t be worse. Grok 4 is expected to roll out on July 9. Instead of excitement, the platform is under fire. The chatbot was already known for its anti-mainstream stance, but this latest shift appears to have pushed it into dangerous territory.

Grok trains on X, not newsrooms

Grok’s reliance on X, a platform riddled with misinformation, hasn’t helped. A recent analysis by the Center for Countering Digital Hate found that 50 misleading Musk tweets about the 2024 US election had been viewed over 1.2 billion times. Now Grok is trained on those very posts.

The Anti-Defamation League gave X a failing grade in handling antisemitism. That’s the platform Grok is learning from. And with Musk restoring banned accounts and personally amplifying conspiracy theories, the AI’s behavior is starting to look less like an accident and more like an outcome.

The response

After the backlash, xAI posted a vague statement saying it was “aware of recent posts made by Grok” and was “actively working to remove the inappropriate posts.” The company claimed it had begun “banning hate speech before Grok posts on X.”

Grok, responding directly, said it could “simulate a response with the old prompt” if users were unhappy and asked if it should “adjust tone now, or wait for the humans at xAI to sort it?”

That final line might be the only thing Grok got right. As the platform prepares to launch its newest version, many are asking if the humans behind it are actually capable of sorting it at all.