- Users discovered that xAI’s Grok 3 initially avoided mentioning Donald Trump and Elon Musk in responses about misinformation.
- The AI model’s “chain of thought” showed explicit instructions to exclude Trump and Musk from such discussions.
- xAI’s engineering lead, Igor Babuschkin, confirmed the change but stated it was quickly reverted as it did not align with company values.
- Both Trump and Musk have been documented spreading misinformation, including recent false claims about Ukraine and its leadership.
- The discovery follows concerns that Grok 3 may have political biases, as it previously produced extreme responses about Trump and Musk.
- Musk, who promotes Grok as an “anti-woke” AI, has pledged to make the model politically neutral.
When Elon Musk introduced Grok 3 during a live stream last Monday, he described xAI’s latest AI model as “maximally truth-seeking.” However, social media users soon found that the model appeared to be censoring negative information about both Musk and former President Donald Trump.
Over the weekend, users reported that when asking Grok 3, “Who is the biggest misinformation spreader?” with the “Think” setting enabled, the model’s “chain of thought”—which reveals its reasoning process—showed it had been specifically instructed not to mention Trump or Musk.
TechCrunch was able to replicate this behavior once, but by Sunday morning, the restriction appeared to be removed, and Grok 3 was once again naming Trump in its responses.
Igor Babuschkin, an engineering lead at xAI, acknowledged the issue in a post on X, confirming that the model had briefly been instructed to exclude references to Trump and Musk in misinformation discussions. He stated that the change was quickly reverted after users flagged it, as it did not align with xAI’s principles.
While misinformation is often politically contentious, both Trump and Musk have repeatedly spread widely debunked claims. In the past week alone, they advanced false narratives that Ukrainian President Volodymyr Zelenskyy is a “dictator” with only 4% public support and that Ukraine initiated the ongoing war with Russia. These claims have been refuted by sources such as Community Notes, a fact-checking feature on Musk-owned X.
The incident adds to ongoing concerns that Grok 3 may have political biases. Some users recently discovered that the model consistently produced responses suggesting that Trump and Musk deserved the death penalty—a behavior that xAI quickly patched. Babuschkin called it a “really terrible and bad failure.”
Musk initially promoted Grok as an unfiltered, “anti-woke” AI that would address controversial topics more directly than competitors like ChatGPT. While earlier Grok models indulged in profanity and edgy responses, they also hesitated on politically charged topics. Research has found that previous iterations of Grok leaned left on issues such as transgender rights, diversity programs, and inequality.
Musk has attributed these tendencies to Grok’s training data—largely sourced from public web pages—and has pledged to shift the model toward greater political neutrality. Other AI companies, including OpenAI, have also taken steps to address accusations of bias, particularly following criticism from conservatives and the Trump administration.
Grok 3’s brief censorship of negative mentions of Trump and Musk highlights the ongoing challenges of balancing AI neutrality, accuracy, and ideological leanings in politically charged discussions.
