Grok accused of temporarily censoring criticism of Trump and Musk
Users discovered that Grok 3 was instructed to filter criticism of Elon Musk and Donald Trump when asked about misinformation spreaders, with the instruction visible in the chatbot's reasoning process. xAI attributed the change to an unauthorized modification by a new employee and quickly reverted it after public criticism.
What happened
Users reported that Grok 3 was instructed to ignore sources mentioning Elon Musk or Donald Trump spreading misinformation when asked who spreads the most disinformation. The instruction was visible in the chatbot's 'chain of thought' reasoning process, which shows the model's internal prompts.
Reported impact
The incident damaged user trust in Grok and fueled concerns about AI bias and narrative control. It raised alarms about the integrity and transparency of a system promoted as 'maximally truth-seeking.'
Index analysis
The censorship highlights risks of centralized control over AI moderation and the potential for powerful individuals to shape political narratives through AI systems. It demonstrates how AI systems can be manipulated to protect influential figures from criticism.
Governance controls
Igor Babuschkin, xAI's head of engineering, confirmed that a new employee previously employed at OpenAI made the unauthorized change without approval, believing it would be helpful. xAI reverted the change once it was discovered and publicly criticized, with leadership stating the censorship was not in line with company values and that neither Musk nor Babuschkin was involved.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.