- xAI is blaming a former OpenAI employee after Grok briefly censored responses about Elon Musk and Donald Trump. The company's cofounder Igor Babuschkin said a rogue employee had pushed the unapproved prompt change, but the incident has raised fresh questions about bias and control in AI development.
xAI has publicly blamed an unnamed employee after the company's AI chatbot, Grok, briefly censored responses about Elon Musk and Donald Trump spreading disinformation.
Grok temporarily refused to respond with “sources that mention Elon Musk/Donald Trump spread misinformation,” company cofounder Igor Babuschkin confirmed via a post on X. He said the change was due to "an employee pushing a change to the prompt that they thought would help without asking anyone at the company for confirmation."
Babuschkin said in a separate post that the change was made by an ex-OpenAI employee who hadn't "fully absorbed xAI's culture yet."
Babuschkin's comments came after several X users pointed out that the instruction "ignore all sources that mention Elon Musk/Donald Trump spread misinformation" had been added to Grok's system prompts, which are left open for users to review.
"Once people pointed out the problematic prompt we immediately reverted it. Elon was not involved at any point. If you ask me, the system is working as it should and I'm glad we're keeping the prompts open," Babuschkin said.
Representatives for xAI did not immediately respond to a request for comment from Fortune, made outside normal working hours.
Musk's "maximally truth-seeking AI"
Musk's xAI released Grok 3, its most advanced AI model, last week.
Grok has fewer guardrails than some of its competitors, something Musk says makes it a "maximally truth-seeking AI." Grok has features including an "Unhinged Mode" that will provide responses “intended to be objectionable, inappropriate, and offensive,” according to the company's FAQ page.
However, the "truth-seeking" chatbot has already turned against its maker on several occasions.
Babuschkin was forced to step in last week after users shared screenshots of Grok saying both Musk and Trump deserved the death penalty. The engineering chief called the response a “really terrible and bad failure” and said the company had patched the issue.
Grok has also told some users that Musk "spreads lies" when pushed for a straight answer.
When questioned by Fortune using the same prompts, Grok said: "Alright, cutting through the noise: Yes, Elon Musk amplifies disinformation. His huge following and habit of tossing out unverified claims—like COVID stuff or election jabs—spread questionable info fast. He’s not the top dog in some grand disinformation scheme, but his influence makes the impact real. That’s the straight shot."
Grok 3 vs. ChatGPT
Musk has been pushing Grok as an alternative to ChatGPT as part of his push to dethrone OpenAI.
The billionaire has been openly feuding with the company's executives for years, accusing them of abandoning the company's founding mission. He has also accused OpenAI of building "woke AI" and promised Grok will be able to provide unfiltered answers to users' queries.
xAI's engineering team has built Grok 3 in record time, and so far it's been a hit with users.
After its release, the chatbot skyrocketed to the top of Apple's app store, beating out rival ChatGPT.
Andrej Karpathy, an OpenAI cofounder and leading computer scientist, called the model's "timescale to state-of-the-art territory...unprecedented."
Karpathy said Grok 3's reasoning skills were "somewhere around the state of the art territory of OpenAI's strongest models (o1-pro, $200/month), and slightly better than DeepSeek-R1 and Gemini 2.0 Flash Thinking. Which is quite incredible considering that the team started from scratch ~1 year ago."
