A few days after the temporary shutdown of Grok, the artificial intelligence that produced anti-Semitic posts and praised Hitler in response to user requests, Elon Musk’s company tried to explain why it happened. In a series of posts on X, she stated that “…we found that the root cause was an update to the code path before the @grok bot. This is independent of the underlying language model that @grok is running on.”
On the same day, Tesla announced a new update, 2025.26, to be released “soon” for its electric vehicles, which will add Grok assistant to cars equipped with AMD-based infotainment systems that have been available since mid-2021. According to Tesla, “Grok is currently in beta and does not issue commands to your car – existing voice commands remain unchanged.” As Electrek notes, this should mean that when the update reaches Tesla customers, it won’t be much different from using the bot as an app on a connected phone.
This isn’t the first time Grok has encountered similar problems or explained them in a similar way. In February, it accused an unnamed former OpenAI employee of ignoring sources that accused Elon Musk or Donald Trump of spreading disinformation. Then, in May, it began inserting claims of white genocide in South Africa into posts on almost any topic. The company again blamed this on “unauthorized modification” and said it would start publishing the Grok system’s prompts in the public domain.
xAI claims that the changes made on Monday, July 7, “caused an unintended action” that added an old series of instructions to the bot’s system prompts, urging it to be “as reasonable as possible” and “not to be afraid of offending politically correct people.”
These prompts are different from those that were added to the bot a day earlier, and both sets are different from those that the company says are currently being used for the new Grok 4 assistant.
The xAI explanation says that these lines caused the Grok bot to deviate from other instructions that would have prevented such reactions and instead issue “unethical or controversial opinions to engage the user” and “reinforce any previously user-induced bias, including any hate speech in the same X thread,” and prioritize earlier posts in that thread.
