Grok 3 appears to have briefly censored unflattering mentions of Trump and Musk
Elon Musk introduced Grok 3, calling it a "maximally truth-seeking AI." But reports suggest it briefly censored unflattering mentions of Trump and Musk.
You are viewing a single comment's thread from:
link
Users found that when asked about misinformation, Grok 3 avoided mentioning Trump and Musk. It seemed to be explicitly instructed not to name them.
TechCrunch confirmed this behavior but noted that by Sunday morning, Grok 3 had resumed mentioning Trump when asked about misinformation.
Both Trump and Musk have previously spread false claims, including recent misinformation about Ukraine’s president and the war with Russia.
Grok 3 faced criticism for appearing too left-leaning. Some users found it even suggested that Trump and Musk deserved the death penalty.
xAI quickly fixed the issue, with its head of engineering calling it a "really terrible and bad failure."
Musk initially pitched Grok as an AI willing to tackle controversial topics and avoid so-called "woke" censorship.
Previous versions of Grok hesitated on political topics, and one study found that it leaned left on issues like diversity and transgender rights.
Musk blamed this bias on Grok’s training data—public web pages—and promised to shift the model toward political neutrality.
Other AI companies, including OpenAI, have also been adjusting their models, possibly in response to political pressure from the Trump administration.