Trump Admin Signs Deal to Use Elon Musk’s Racist Grok AI Chatbot

The Trump administration on Thursday announced an agreement with Elon Musk’s company xAI, making his questionable chatbot, Grok, available to every federal agency over the next year and a half for a nominal fee of 42 cents.

“We value xAI for partnering with —and dedicating engineers—to accelerate the adoption of Grok to transform government operations,” said Josh Gruenbaum, commissioner of the Federal Acquisition Service, which handles government purchasing. Musk, who publicly feuded with President Donald Trump just a few months ago, expressed excitement about working with his administration “to rapidly deploy AI throughout the government for the benefit of the country.”

Grok’s potential integration within federal agencies is certainly a high point for a chatbot that, in its time as a feature on Musk’s social media platform X, has had some notorious low ones.

There was the period when it began producing antisemitic screeds and calling itself “MechaHitler.” Or when it randomly came to share its South African creator’s obsession with alleged “white genocide” in South Africa—so much so that it would bring it up in unrelated inquiries. Or when it made disturbing sexual comments about then-CEO of X Linda Yaccarino, or similarly harassed prominent liberal X user Will Stancil, providing users instructions to break into his house, and reminding them to bring “lockpicks, gloves, flashlight, and lube—just in case.”

Musk, for his part, has made comments raising concerns about his possible tinkering with Grok such that it shares his right-wing beliefs. When Grok told a user that “right-wing political violence has been more frequent and deadly” than left-wing violence since 2016, while providing evidence, Musk called it a “major fail,” accusing the chatbot of “parroting legacy media.”

Days later, he announced Grok 4, the model available to federal agencies under the new xAI-GSA agreement. He promised to train Grok 4 by having the previous version of the chatbot “rewrite the entire corpus of human knowledge, adding missing information and deleting errors. Then retrain on that.” He also asked X users to provide “divisive facts” on which to train the model—“things that are politically incorrect, but nonetheless factually true”—which elicited replies such as “Covid-19 Vaccines did not save a single life” and “Nonbinary isn’t a real thing.”