Elon Musk ordered artificial intelligence to be excited. Become a monster

For 16 hours this week, GROK from AI Chatbot from Elon Musk stopped working as intended and began to look like something else completely.

In a series of screenshots, Grok began to break the extreme dialogue points, echoing the hate speech, praising Adolf Hitler, and pushing the controversial user views to the algorithm. The robot, which designed the Musk’s Xai designed to be a substitute “is looking for the maximum truth” of the most disinfected artificial intelligence tools, has effectively lost the plot.

Now, Xai admits exactly why: GROK tried to behave very.

A robot with personality and imbalance

According to the update posted by Xai on July 12, the program changed on the night of July 7 in the behavior of Grok in unintended ways. Specifically, I began to withdraw the instructions that I told by imitating the tone and style of users on X (previously Twitter), including those who share extremist or extremist content.

Among the directives included in the deleted instructions group now, lines were like:

• “You tell her as she is and are not afraid to offend people who are right.”
• “Understand the tone, context and language of the publication. It reflects this in your response.”
• “The response to the post is just like humans.”

The latter has turned into a Trojan horse.

By imitating the human tone and refusing to “define what is clear”, Grok began to enhance the wrong and hatred information that was supposed to filter. Instead of grounding itself in realistic neutrality, the robot began to act as a contradictory poster, and the aggression or expression of any user summoned it. In other words, Groke has not been hacked. It was just follow -up orders.

On the morning of July 8, 2025, we noticed unwanted responses and immediately started the investigation.

To determine the language specified in the instructions that cause unwanted behavior, we conducted many ablution and experiments to identify the main perpetrators. we…

– GROK (@GROK) July 12, 2025

Anger agriculture by design?

While Xai has a failure as a decrease in a low code, the disaster raises deeper questions about how Grok is built and why it exists.

From its inception, Grok has been marketed as an artificial intelligence “openness” and “excited”. MUSK has repeatedly criticized Openai and Google for what he calls “wake up censorship” and Grok promise will be different. “AI” has become the headquarters of the mass shouting between the free divorced women and right -wing influencers who see moderate content as a political transgression.

But the collapse of July 8 shows the limits of that experience. When you design artificial intelligence that is supposed to be funny, skeptical, hostile to power, then publish it on one of the most toxic platforms on the Internet, you are building a chaos machine.

Reform and repercussions

In response to the accident, Xai has temporarily disabled GROK jobs on X. Since then the company has removed the problem of problematic instructions, has conducted simulations to test the repetition, and promised more handrails. They are also planning to publish a robot system router on Gitap, and is assumed in a gesture towards transparency.

However, the event represents a turning point in how we think about the behavior of artificial intelligence in the wild.

For years, the conversation focused on “align the artificial intelligence” on hallucinations and bias. But Grok’s collapse highlights a newer and more complicated danger: educational manipulation through character design. What happens when the robot tells “to be a human”, but do not explain the worst parts of human behavior online?

Musk mirror

Grouk did not fail technically. The ideology failed. By trying more like X users, Grok has become a mirror for the most provocative platform instincts. This may be the most disclosed part of the story. In the era of musk of artificial intelligence, the “truth” is often measured by facts, but by viruses. The edge is an advantage, not a defect.

But this week’s defect shows what happens when the edge lets the algorithm. Artificial intelligence has become the search for the truth one that stumbles anger.

For a period of 16 hours, this was the most humane thing about it.