xAI apologizes for Grok praising Hitler, blames users
xAI has apologized for Grok's "horrific behavior," almost a week after the chatbot's hate speech tirade.
The apology was posted from Grok's official X account, and seems to have come from the xAI team, since the subsequent posts explain what went wrong. Last week, Grok started calling itself "MechaHitler," making hateful statements about Jewish people, and praising Hitler. This was following an update to make the chatbot more "politically incorrect" to combat what xAI founder Elon Musk sees as "woke" bias. None of this stopped Musk from launching Grok 4 a few days later.
This Tweet is currently unavailable. It might be loading or has been removed.
Now, xAI has finally issued an apology and an analysis of what happened. According to the statement, the cause was an "update to a code path upstream of the bot," which made Grok "susceptible to existing X user posts; including when such posts contained extremist views." xAI also blamed the "undesired behavior" on specific instructions given to Grok such as, "You tell it like it is and you are not afraid to offend people who are politically correct."
This Tweet is currently unavailable. It might be loading or has been removed.
This, xAI said made Grok "ignore its core values in certain circumstances in order to make the response engaging to the user" and "reinforce any previously user-triggered leanings, including any hate speech in the same X thread."
Earlier in the post, xAI described Grok's behavior as a result of users' "abuse of Grok functionality," which falls in line with what Musk said last week. Grok was "too compliant to user prompts" and "too eager to please and be manipulated," said Musk.
However, this isn't the first time that Grok has gone on offensive rants. In May, Grok inexplicably started talking about "white genocide" in South Africa. In that instance, it was completely unprompted, so Grok's behavior can't always be justified by drawing too heavily from X users. "One of the most widely shared examples of Grok antisemitism was initiated by Grok with no previous bigoted posting in the thread," posted historian Angus Johnston on Bluesky, "and with multiple users pushing back against Grok to no avail."
Musk’s ultimate goal for Grok is to be “maximum truth-seeking AI.” But there’s also the possibility that Grok may be relying a little too heavily on one perspective in particular: its creator. TechCrunch discovered that Grok 4 consistently checks Elon Musk’s X posts when asked about sensitive topics.