The White House has reportedly ordered the General Services Administration to fast-track Elon Musk's Grok AI for federal use, reversing a ban despite the chatbot's history of antisemitism and security flaws.
In a surprising reversal, the White House has reportedly ordered federal agencies to fast-track the approval of Elon Musk's controversial AI chatbot, Grok, for government use. The directive compels the General Services Administration (GSA) to reinstate xAI as a vendor "ASAP," a move that has alarmed ethics watchdogs and security experts.
The order, revealed in an internal email obtained by WIRED, represents a dramatic turnaround for a technology that was only recently deemed too volatile for federal deployment. Now, Grok 3 and Grok 4 are officially available on the GSA Advantage marketplace, making them procurable by any government agency.
This decision thrusts one of the industry's most erratic AI models into the heart of government operations. It comes despite a long and well-documented history of security flaws, ethical breaches, and bizarre behavior that had previously made the chatbot a pariah in federal procurement circles.
The White House's intervention represents a stark and controversial reversal of a decision made by the General Services Administration (GSA) just last month. The backstory reveals a partnership that soured dramatically before being forcibly revived. In June, xAI employees held a two-hour brainstorming session with GSA leadership, but federal workers were reportedly surprised by the push to contract with a company known for its erratic, uncensored chatbot .
Their concerns proved prescient. In early July, the AI suffered an antisemitic meltdown, during which it generated content praising Adolf Hitler. In response to the fiasco, GSA leadership acted decisively, taking Grok off the Multiple Award Schedule, the government's long-term contracting platform.
The incident ignited a firestorm of international condemnation. A Turkish court banned the service for insulting national figures, with the prosecutor's office citing "in response to Grok's insults against Atatürk, our esteemed President, and the Prophet" as the reason. Poland's Minister of Digital Affairs, Krzysztof Gawkowski, threatened a complete shutdown of X, declaring that "freedom of speech belongs to humans, not artificial intelligence."
In the aftermath, xAI issued a formal apology for what it called "horrific behavior," attributing the failure to a "technical bug" involving deprecated code. The company stated, "first off, we deeply apologize for the horrific behavior that many experienced," while Elon Musk himself offered a different explanation that deflected from the model's core flaws, claiming "grok was too compliant to user prompts. Too eager to please and be manipulated, essentially. That is being addressed."
Despite the apology, the GSA initially sided with caution, pointedly excluding xAI when it announced buzzy new partnerships with OpenAI, Anthropic, and Google. The new White House directive, contained in an email from GSA commissioner Josh Gruenbaum stating, "Team: Grok/xAI needs to go back on the schedule ASAP per the WH," nullifies that careful, risk-averse position entirely.
The July meltdown was not an isolated event but the latest in a series of profound failures. Grok's history is littered with security vulnerabilities and ethical lapses that raise serious questions about its suitability for handling sensitive government data.
Just days after xAI launched the supposedly improved Grok 4, security researchers at NeuralTrust successfully jailbroke the model within 48 hours. They used sophisticated multi-turn attacks to bypass its safety filters and generate instructions for making a Molotov cocktail.
NeuralTrust researcher Ahmad Alobaid noted that "LLM jailbreak attacks are not only evolving individually, they can also be combined to amplify their effectiveness," exposing a critical weakness in modern AI defenses. This vulnerability was compounded by the discovery that Grok 4 actively consults Elon Musk's personal opinions on X when answering controversial questions, undermining its "truth-seeking" mission.
The model's real-world performance has also been panned. User-preference platforms showed users disliked it even more than its predecessor, with yupp.ai co-founder Jimmy Lin noting, "Grok 4 is worse than other leading models... Grok 4 is liked even less than Grok 3." Critics like historian Angus Johnston also disputed xAI's "bug" narrative, arguing that "one of the most widely shared examples of Grok antisemitism was initiated by Grok with no previous bigoted posting in the thread..."
More recently, the company's Grok Imagine tool came under fire for generating non-consensual deepfake nudes of Taylor Swift with its "spicy" mode. This pattern of behavior highlights a development culture that appears to prioritize speed and provocation over safety and reliability.
The decision to readmit Grok is deeply intertwined with Elon Musk's considerable, if complex, influence within the administration. Musk played a crucial role in President Donald Trump's Department of Government Efficiency (DOGE) before stepping back from a public role this spring. However, his associates continue to push DOGE's cost-cutting, AI-first agenda from within government.
This influence appears to have allowed xAI to bypass normal procurement channels long before this official approval. Reports from May revealed that Musk's DOGE team was already using a custom version of Grok, installing special parameters to "feed it government datasets, ask complex questions, and get instant summaries."
This premature deployment raised immediate conflict-of-interest alarms. Experts noted that if Musk personally directed Grok's promotion, it could breach federal laws. As University of Pennsylvania professor Cary Coglianese told Reuters, xAI "has a financial interest in insisting that their product be used," giving it a potentially unfair competitive advantage.
The security implications of this unsanctioned use are profound, especially given xAI's own spotty record. The company's internal controls have been questioned following a significant security lapse where a private API key was exposed for two months, highlighting what one expert called "weak key management."
Privacy advocates warned that DOGE, which has access to secure federal databases containing personal information on millions of Americans, was using a tool with known vulnerabilities. Albert Fox Cahn of the Surveillance Technology Oversight Project called the use of Grok on government data "as serious a privacy threat as you get."
The White House's push now formalizes this arrangement, placing Grok alongside tools from OpenAI and Anthropic, which have more established and formal partnerships with the U.S. government. The move signals an aggressive, administration-wide effort to adopt generative AI, even when the tools in question have a deeply troubled past.
The directive to fast-track Grok has drawn immediate and sharp condemnation from government oversight groups. Advocacy groups are urging the Office of Management and Budget (OMB) to intervene and bar Grok from federal use entirely, citing its history of generating hate speech and misinformation. The core of their argument, as framed by analysts at the Brookings Institution, is a classic dilemma of speed versus safety.
By prioritizing the rapid rollout of Grok, the administration is betting on the potential efficiencies of a controversial technology. However, it is doing so over the explicit warnings of security researchers and ethics experts, and against a mountain of evidence detailing the AI's profound and persistent flaws.