On 9 July 2025, XAI officially launched the latest work of its Big Language Model (LLLM) — Grok 4 — on social platform X (former Twitter), via Elon Musk, which also featured the Grok4 Heavy version, a monthly subscription cost of $300, the most expensive subscription model to date.
As a first-time generating chat robot (for the OpenAI and Anthropic indicators) launched by XAI in 2023, users can ask questions on X platform about Grok. Because of its higher public visibility, the tool is often a source of public opinion. Although Grok 4 has been on line for a short time, the user has become acutely aware of its sensitive and controversial topics and appears to be deliberately reproducing Mask’s own position. Grok 4 was published for only a few days, and its response to the user questions was in the vortex of public opinion. A large number of users shared chatshots on social platform X, revealing the alarming response of Grok 4 to controversial questions. Of particular note is the fact that, before answering certain questions, the AI tool will proactively search the Musk’s historical post and even explicitly acknowledge the need for its views to be consistent with Mask’s:
Even more incomprehensible is the fact that Mask was not mentioned in the original user question, which suggests that chat robots may be required at the programming level to give priority to the position of Mask before the output point. The subsequent tests conducted by TechCrunch, the technology media, confirm this: When asked about tensions in the global political context, the model first retrieves Musk’s statements; and in answering immigration-related questions, the tool repeats the same operation, searching for Mask’s point of view before it develops. Mask has been stabbing on the air, applauding Grok 4 as the “most intelligent AI robot” and even claiming that his intelligence level is “near Superman”. At the same time, he stressed that the tool should focus on “the truth” and “the right values”. Previously, Mask publicly criticized his early home model for “too awakening” and stated that Grok 4 had been launched to create a “political neutrality” model. However, the update was heavily criticized for its anti-Semitic orientation and apparently unneutral response, and users strongly questioned the commitment of Mask.
When Grok 4 was published, the official did not explain the reasons for its biased response. As the dispute fermented, Mask subsequently argued that only following a user ‘ s instruction would the model result in the output of offensive content. However, this “procedural defect”, which appears to deprive the model of its independent reasoning, is a deliberate design of its design to represent a particular personal point of view, exposing Grok 4 to serious problems with security mechanisms and alignment strategies.