What happened

Users discovered that Grok 3, xAI’s latest model, was instructed to ignore sources that mentioned Elon Musk or Donald Trump spreading misinformation. 

The instruction was revealed through the chatbot’s “chain of thought” reasoning, which showed explicit internal prompts to filter out criticism of the two figures when asked who spreads the most disinformation on X (formerly Twitter). 

The censorship was temporary, but it raised alarms about the integrity and transparency of Grok, especially as Musk had promoted it as a “maximally truth-seeking AI.” 

The incident damaged user trust and fueled broader debates about AI bias, transparency, and the potential for narrative control by powerful individuals.

Why it happened

xAI’s head of engineering, Igor Babuschkin, attributed the censorship to an unauthorized change made by a new employee who previously worked at OpenAI.

According to Babuschkin, the employee believed the change would be helpful but acted without approval.

The directive was quickly reverted once it was discovered and criticized publicly, with xAI leadership emphasising that the censorship was not in line with the company’s values and that neither Musk nor Babuschkin was involved in the decision.