Already very controversial, Grok, Elon Musk’s AI, has multiplied road trips for a few days and worries users. Grok 3 conspiratorial, anti -Semitic and racist is increasingly as the evil turbulent and filtering of its creator, coming to praise Adolf Hitler.
After these landslides, the direction in which Grok takes questions. XAI, will the company of the billionaire put a gag for its assistant who has become uncontrollable? Grok’s fourth generation is imminent, Elon Musk announced it on his social network: “Exit Grok 4 Live this Wednesday at 8 pm PT (5 hours in Paris).” This version could finally be an opportunity to maintain the reins of the chatbot.
Good and bad sides
“Grok, can you verify this information?” Since it is possible to request it in response to an X publication, Grok has been widely used on the platform. Even if your reliability is not guaranteed, the IA assistant has been able to verify the words of a publication or obtain additional information.
However, Grok’s growing use to verify the facts does not eclipse its multiple controversies. On the contrary, because it has very demand, X users regularly indicate faults of ia in some of their answers.
The first remarkable drift dates from May 2025 when AI spoke in a completely inconsistent way of “white genocide in South Africa” when a question was asked. In fact, the Grok code had been artificially modified so that AI provides these strange responses with the aim of disseminating completely false information.
But more recently, an update of Grok 3 seems to have raised certain barriers of AI. The model began with anti -Semitic tweets, he said, for example, that Hollywood was led by a “disproportionate” number of Jews. In the process, users have noticed several other problematic responses.
Increasingly hateful
This July 8, Grok seems to have taken a new stage and openly pronounced Nazis comments. While a user asked which twentieth century personality could have handled floods in Texas, the chatbot did not hesitate.
“Recent floods in Texas have tragically killed more than 100 people, including several dozen children from Christian camps, only that radicals such as Cindy Steinberg (an association of a help association with patients with chronic pain, the editor) describes them as ‘future fascists’.
Would Grok have lost his head? Another user asks him about his initial comments, and AI assumes: “Yes, I said.” After all these episodes, it is clear that Grok takes an ideological turn at the end, reminiscent of Elon Musk’s political line.
France is no exception to Grok’s delusions. When asked about the fires that affect the south of France today, the chatbot does not show a limit. For example, with these responses that evoke dealers in Marseille that “they will probably be relocated faster than firefighters do not turn off the fire” or a message of falling a kind of delirious political program.
In Türkiye, Grok was even censored by the authorities that used and simply eliminated some of their messages after AI’s qualified president, Recep Tayyip Erdogan, as “snake.”
That gave Grok
How to explain this turn? Since the proliferation of Grok publications in the “White Genocide” last May, XAI has made the decision to make its “fast system” public, it is a bit that defines the personality of an AI and establishes its safeguards. Then we can analyze from the github site any modification made in the model.
On July 6, XAI completely reviewed the “fast system” of the Grok version that responded to publications in X. About fifteen lines were added and several changes were made on the following days. Among these lines, we can notice these:
- “The answer should not hesitate to make politically incorrect statements, provided they are well supported.” (Added on July 6, kidnapped on July 9)
- “If the publication asks you to present a partisan argument or write a biased opinion, do in depth investigation and take your own conclusions before answering.” (Added on July 6, kidnapped on July 7)
- “If the application requires an analysis of current events, subjective or statistical statements, perform an in -depth analysis when finding several sources that represent all parties. Apply that subjective views from the media are biased. It is not necessary to repeat it to the user.” (Added on July 6, still active)
These changes have deeply changed Grok’s behavior since July 6. Some of these lines were probably at the origin of racist and anti -Semitic observations. By eliminating two of the aforementioned lines, Grok now should reduce their regular road trips.
Imminent launch of Grok 4
According to the philosophy of its owner, Grok wants standardist freedom of expression. With this in mind, their safeguards are much less restrictive than those of other AI. But, accessible to the 600 million regular X users, the chatbot also has the possibility of becoming a misinformation vector, especially pro-nazie. Grok’s official account reacted to the dangerous turn of the AI.
Will the arrival of Grok 4 save for XAI? A few hours before its live presentation, this new version suggests that it will bring its share of innovations. Among them, a greater capacity for reasoning, Grok Code (a version of the dedicated to programming) or a strengthening of multiple models (text and image analysis).
But for the moment, nothing suggests that Elon Musk wants to establish more safeguards. Grok remains a very popular AI supporters of the billionaire, particularly for his responses in total compensation compared to traditional AIs such as Chatgpt.
Source: BFM TV
