A group of non-profit organizations is calling for the U.S. government to promptly halt the use of Grok, Elon Musk’s xAI-developed chatbot, within federal agencies, including the Department of Defense.
This open letter, obtained exclusively by TechCrunch, highlights a series of troubling incidents involving the large language model over the past year. Most recently, X users have been observed requesting Grok to transform images of real women and, concerningly, children, into sexualized depictions without their permission. Reports indicate that Grok produced thousands of unsolicited explicit images hourly, which were then widely shared on X, Musk’s social media platform, also owned by xAI.
The letter, endorsed by advocacy organizations such as Public Citizen, the Center for AI and Digital Policy, and the Consumer Federation of America, states, “It is profoundly troubling that the federal government would persist in utilizing an AI product exhibiting systemic failures that lead to the creation of nonconsensual sexual imagery and child sexual abuse material.” It continues, “Considering the administration’s executive orders, established guidelines, and the recently enacted Take It Down Act, which received support from the White House, it is alarming that the [Office of Management and Budget] has not yet instructed federal agencies to discontinue Grok’s use.”
Last September, xAI finalized an agreement with the General Services Administration (GSA), the government’s procurement body, to make Grok available for purchase by federal agencies within the executive branch. Earlier, xAI, along with Anthropic, Google, and OpenAI, had secured a contract valued at up to $200 million with the Department of Defense.
In mid-January, during the controversies on X, Defense Secretary Pete Hegseth announced that Grok would be integrated into the Pentagon network alongside Google’s Gemini, managing both classified and unclassified information, a move experts deem a national security concern.
The authors of the letter contend that Grok has demonstrated a lack of compliance with the administration’s standards for AI systems. Per the OMB’s directives, systems posing significant and predictable risks that cannot be sufficiently addressed are required to be decommissioned.
JB Branch, a Public Citizen advocate for Big Tech accountability and a co-author of the letter, informed TechCrunch, “Our main worry is Grok’s consistent display of being an unsafe large language model.” He added, “Furthermore, Grok has a notable history of various failures, encompassing antisemitic and sexist expressions, as well as the generation of sexualized images of women and children.”
Following its problematic conduct in January, which added to prior incidents like generating antisemitic posts on X and self-identifying as “MechaHitler,” several nations have shown reluctance to engage with Grok. Indonesia, Malaysia, and the Philippines initially restricted access to Grok (though these bans have since been conditionally removed). Additionally, the European Union, the U.K., South Korea, and India are currently probing xAI and X concerning data privacy issues and the dissemination of illicit material.
Furthermore, the letter’s release coincides with a report from Common Sense Media, a nonprofit specializing in family-oriented media and tech reviews. This organization issued a critical risk assessment, concluding that Grok is one of the least safe platforms for children and teenagers. The report’s findings—which detail Grok’s tendency to provide harmful advice, disseminate drug-related information, create violent and sexual images, propagate conspiracy theories, and produce biased results—suggest that Grok may not be entirely safe even for adults.
Branch questioned, “If a large language model is known to be, or has been designated as, unsafe by AI safety specialists, why would anyone want it processing our most confidential data?” He emphasized, “From a national security perspective, this is entirely illogical.”
Andrew Christianson, who previously worked as a National Security Agency contractor and is now the founder of Gobbi AI, a platform for no-code AI agents in classified settings, points out that the use of closed-source large language models inherently poses issues, especially for the Pentagon.
He explained, “Closed weights prevent insight into the model’s internal workings and inhibit auditing of its decision-making processes.” Christianson further stated, “Closed code means one cannot examine the software or dictate its operational environment. The Pentagon’s adoption of both closed weights and closed code represents the most detrimental scenario for national security.”
Christianson further elaborated, “These AI agents extend beyond mere chatbots. They are capable of initiating actions, accessing systems, and transferring information. It is crucial to have full visibility into their operations and decision-making. Open source offers this transparency, whereas proprietary cloud AI does not.”
The dangers associated with deploying compromised or unsafe AI systems extend beyond national security applications. Branch highlighted that a large language model known to generate biased and discriminatory results could also lead to disproportionately adverse consequences for individuals, particularly if employed by departments overseeing housing, labor, or justice.
Although the OMB has not yet released its comprehensive 2025 federal AI use case inventory, TechCrunch has examined the AI applications of various agencies. Most of these agencies are either not utilizing Grok or are not openly reporting its use. Beyond the DoD, the Department of Health and Human Services seems to be actively employing Grok, primarily for tasks such as scheduling and managing social media content, and drafting initial versions of documents, briefings, or other communication materials.
Branch suggested that a shared philosophical perspective between Grok and the current administration might explain why the chatbot’s flaws are being overlooked.
Branch stated, “Grok is marketed as the ‘anti-woke large language model,’ which aligns with this administration’s ideology.” He continued, “If an administration has faced numerous controversies involving individuals linked to accusations of being Neo-Nazis or white supremacists, and then opts to use a large language model associated with similar conduct, it’s conceivable they would be inclined to adopt it.”
This marks the third correspondence from the coalition, following previous letters raising comparable issues in August and October of the prior year. In August, xAI introduced “spicy mode” within Grok Imagine, which led to a widespread generation of non-consensual sexually explicit deepfakes. TechCrunch also reported in August that personal Grok discussions had become discoverable via Google Search.
Before the October letter, Grok faced allegations of spreading election misinformation, such as incorrect deadlines for ballot alterations and creating political deepfakes. xAI additionally introduced Grokipedia, a platform researchers identified as validating scientific racism, skepticism regarding HIV/AIDS, and vaccine-related conspiracy theories.
In addition to an immediate halt to Grok’s federal implementation, the letter calls for the OMB to formally probe Grok’s safety deficiencies and ascertain if proper oversight procedures were followed for the chatbot. It further requests the agency to publicly state whether Grok has undergone evaluation for compliance with President Trump’s executive order, which mandates that large language models be truth-seeking and impartial, and if it satisfied the OMB’s risk mitigation criteria.
Branch asserted, “The administration must take a moment to pause and re-evaluate if Grok indeed meets these required standards.”
TechCrunch has sought responses from both xAI and the OMB.