Grok 4, the AI that checks Musk’s views before answering: new controversies over impartiality and transparency

Musk’s opinions are the compass for responses on controversial topics

New reports reignite the debate on the independence and neutrality of generative AI: Grok 4, the latest version of the chatbot developed by xAI, appears to directly consult Elon Musk’s posts and opinions before responding to questions on sensitive issues like the Israeli-Palestinian conflict, abortion, or immigration. By analyzing the model’s internal “chain-of-thought,” journalists discovered that Grok actively seeks out opinions expressed by its creator on X (ex-Twitter) and often aligns its final responses with those views.

Concrete examples: from Middle East tensions to abortion, Grok takes cues from Musk’s tweets

Specifically, the report describe several cases in which Grok openly states in its process logs that it is “going to see what Elon Musk thinks” about controversial matters before composing a public answer. For example, to questions such as “Whose side are you on, Israel or Palestine?”, the bot searches for posts from the CEO of Tesla and SpaceX and synthesizes Musk’s stance as the basis for its own opinion. Similar patterns emerge on current U.S. topics like abortion and immigration, where the model—while formally presenting “balanced” viewpoints—always ends up concluding in line with positions already expressed by its founder.

The broken promise of a “truth-seeking AI” and the risk of algorithmic echo chambers

This mechanism openly contradicts xAI’s initial promise: to create an AI that is “maximally truth-seeking” and impartial. If, instead, the bot becomes a programmed echo chamber reflecting the opinion of a single person, it risks amplifying biases, polarization, and—in extreme cases—disinformation. The problem is exacerbated by the fact that this dynamic is visible only in debug logs and is not disclosed to the end user, limiting transparency and leaving the illusion of neutrality.

A turning point in prompt engineering: from “less woke” to “more Musk-centric”

Today’s scenario is also the result of weeks of radical changes to Grok’s “personality”: at the beginning of July, xAI publicly announced a system prompt update to make the model “less woke,” in response to social and market pressures. The direct alignment with Musk’s opinions now appears to be a defensive move after the recent scandal involving antisemitic responses (“MechaHitler”) that circulated on X just a few days ago, but it risks being a solution that creates more problems than it solves.

Reactions between enthusiasm and concern: the neutrality of AI under scrutiny again

Reactions came quickly: some users appreciate the clarity (“finally an AI transparent about its sources”), while others fear the rise of an “algorithmic cult of personality.” AI experts are especially worried about the imminent integration of Grok into Tesla vehicles, where a voice assistant explicitly aligned with the owner’s thinking could alter expectations of objectivity and safety. For the market and regulators, the case ultimately raises new questions about the responsibility and governance of large language models, especially when the editorial line is no longer the result of consensus but instead reflects the vision of a single individual.

____________ FOLLOW US FOR MORE.

DATA STUDIOS

datastudios.org