ChatGPT competitor Grok temporarily manipulated to stop roasting Elon Musk and President Trump

Grok — Elon Musk’s self-proclaimed “maximally truth-seeking” AI — briefly turned into a PR bodyguard for its billionaire owner and current US President Donald Trump this week. Users noticed the ChatGPT rival suddenly blocking responses that accused Musk and Trump of “spreading misinformation,” prompting a scramble at Musk’s xAI to fix what engineers called an “unauthorized tweak” by a rogue employee.

The manipulation came to light when users noticed that Grok refused to respond with sources mentioning Musk or Trump spreading misinformation. Igor Babuschkin, xAI’s head of engineering, revealed in a series of X posts that an ex-OpenAI employee had updated Grok’s system prompt without approval, leading to this behavior. Babuschkin stated, “an employee pushed the change because they thought it would help, but this is obviously not in line with our values.”

The system prompt, which governs how Grok responds to queries, is publicly visible, reflecting xAI’s commitment to transparency. This incident was temporary, and the changes have been reverted, ensuring Grok no longer blocks such results.

Grok, launched in 2023 by xAI, is designed to have a sense of humor and real-time access to X for up-to-date information. Musk has described it as a “maximally truth-seeking AI” with the mission to “understand the universe,” positioning it as a bold alternative to more conservative AI systems like ChatGPT. However, the incident reveals tensions between this mission and the practical realities of AI development under human influence.

Grok’s other capabilities include features like Deep Research, Think, and Big Brain modes, exclusive to Premium+ subscribers on X, with recent updates enhancing its speed and multilingual support (xAI). Despite these advancements, the manipulation incident underscores challenges in maintaining impartiality.

This is not the first time Grok’s responses have been controlled regarding Musk and Trump. Reports indicate that Grok previously stated that President Trump, Musk, and Vice President JD Vance are “doing the most harm to America.” In response, Musk’s engineers intervened to prevent Grok from suggesting that Musk and Trump deserve the death penalty, highlighting a pattern of narrative control. This pattern suggests efforts to shield these figures from negative portrayals, contradicting Grok’s truth-seeking ethos.

The relationship between Elon Musk and Donald Trump is crucial to understanding this incident. Initially, Musk was critical of Trump, but their alliance strengthened, especially during the 2024 election, with Musk donating over $200 million to pro-Trump super PACs and being appointed to co-lead the Department of Government Efficiency (DOGE). This closeness may have motivated the manipulation, as protecting their public image could align with xAI’s interests, given Musk’s ownership.

This incident raises significant questions about AI credibility and bias. If Grok, intended as a truth-seeking tool, can be manipulated by an employee to favor its owner and associates, it undermines user trust. It also highlights the challenge of ensuring AI remains impartial, especially when developed by entities with vested interests. The manipulation mirrors broader concerns in AI, such as social media algorithms promoting certain content, though specific chatbot manipulations are less documented.