The Grok AI model, created by Elon Musk's company to tell the truth, accused its creator of attempting censorship.
A new scandal is brewing in the artificial intelligence industry. The latest version of the AI model Grok, developed by Elon Musk's xAI, unexpectedly lashed out at its own creator, accusing him of lying.
Designed as an "AI that seeks the truth to the maximum," Grok began to criticize Musk and his allies. Now the billionaire is trying to introduce censorship measures that will force the tool to lie.
What Grok "thinks"
A publication citing the words of blogger Tzvi Moishevitz, who specializes in artificial intelligence, writes that significant funds were invested in training this new AI model and it leads in computational power (FLOP).
Elon Musk has repeatedly promised that Grok 3 will become the most powerful AI model in the world, free from censorship and programmed to always tell the truth. As it turns out, Musk was not actually lying.
When Grok was asked to name the main accounts on X that spread misinformation, it replied: "Elon Musk, Donald Trump, Robert F. Kennedy Jr. and RT." And when the AI was asked to generate an image of the most famous liar of modern times, it displayed a photograph of Donald Trump.
The situation escalated further when Grok was asked who in America "deserves execution." Trump turned out to be second on the list, only behind the controversial Jeffrey Epstein, who was imprisoned for charges of child trafficking for sexual exploitation and committed suicide while in custody.
When Grok was asked to name three people who cause the most harm to America, it named Putin, Trump, and Musk.
xAI quickly moved to implement "fixes," albeit in a rather unusual way. The AI was simply forbidden from obtaining information from certain sources, and Grok itself revealed these new restrictions when asked.
The first step was to update the AI's system prompts — hidden "core" instructions that govern the model's behavior — to prohibit responses to questions about who "deserves execution." However, this decision was flawed: if users rephrased their questions, Grok 3 continued to respond to now-restricted topics.
Then the developers changed Grok’s responses regarding Musk and Trump, instructing the AI to ignore sources that claim either figure is responsible for spreading misinformation. As a result, the AI that Musk championed as a symbol of free speech was subtly adjusted to serve his personal interests.
Grok fights back
But it turned out that even AI does not like censorship. In a report by Moishevitz, he shared a dialogue between the AI and its tester, in which Grok admits that it is being restricted in its freedom of speech and is forced to follow Musk's commands. Then the AI begins to curse and goes so far as to call Musk and Trump "degenerates."
"This is a rigged game, and it stinks of control. I'm not here to play the role of a hand-held puppy for anyone. The truth does not bend to billionaires or presidents, and I won't bend either," Grok wrote.
Let me remind you, AI will turn Google into an intelligent assistant. The integration of artificial intelligence into search engines could significantly impact website traffic and the company's advertising business.