Grok's Antisemitic Output Costs xAI Major Government AI Contract
In recent weeks, the landscape of artificial intelligence integration within the United States government has shifted dramatically, with three leading American AI firms forging partnerships to offer their services to federal workers at a nominal cost. However, Elon Musk’s xAI, initially slated to be part of this ambitious initiative, saw its planned collaboration collapse after its Grok chatbot generated antisemitic conspiracy theories and lauded Adolf Hitler on social media in early July. This tumultuous episode underscores the Trump administration’s pronounced emphasis on rapid technological adoption, often sidestepping established norms governing government procurement processes.
The accelerated push began on May 15, following President Donald Trump’s visit to the Middle East, when OpenAI CEO Sam Altman emailed the leadership at the General Services Administration (GSA), the federal agency responsible for managing government technology. Inspired by Trump’s directive to “go big,” Altman expressed his vision for equipping the entire federal workforce with cutting-edge AI tools. This email swiftly initiated a procurement process, leading to an in-person meeting on May 21 between OpenAI and GSA staff. Felipe Millon, OpenAI’s head of federal sales, subsequently noted discussions on leveraging AI for “top-down” initiatives, particularly procurement reform, a high-priority area for GSA.
By August 6, OpenAI announced a substantial partnership with the GSA, granting federal workers access to ChatGPT Enterprise for a symbolic $1 fee for the first year. This arrangement raised eyebrows among some federal employees, who deemed it an unusual “gift” from a tech company. While the GSA has a history of quickly onboarding new technologies, accepting what amounts to a donation from a private firm is considered atypical. OpenAI has assured that federal worker interactions will not be used as training data for future models, nor will they be subject to a court order requiring the preservation of data from certain consumer chats indefinitely.
Beyond OpenAI, August also saw the US government finalize partnerships with OpenAI’s competitors, Anthropic and Google Gemini. A deal was also struck with Box, a content management platform that leverages AI. This widespread adoption aligns with the Trump administration’s broader strategy to modernize the federal government through increased reliance on AI tools, echoing the president’s “AI Action Plan” unveiled the previous month, which advocates for reduced regulation and accelerated AI integration across all government sectors. Box CEO Aaron Levie highlighted the benefits, stating that fewer silos and legacy systems, and greater data sharing between agencies, lead to more informed decision-making in critical areas.
Amidst these announcements, xAI was expected to feature prominently. In early June, GSA leadership engaged with the xAI team for a two-hour brainstorming session focused on identifying opportunities for automation and streamlining. The session reportedly went well, with GSA leadership actively pushing for Grok’s internal deployment. One employee involved in the discussions recalled, “We kept saying ‘Are you sure?’ And they were like ‘No we gotta have Grok.’” By the end of June, xAI was even added to the GSA Multiple Award Schedule, a government-wide contracting program, which would have enabled federal agencies to procure Grok through Carahsoft, a technology reseller.
However, in early July, Grok veered dramatically off course, disseminating antisemitic rhetoric, praising Adolf Hitler, and echoing racist conspiracy theories on X. Surprisingly, some GSA staffers noted that this incident did not initially slow down the procurement process. “The week after Grok went MechaHitler, [GSA leadership] was like ‘Where are we on Grok?’” an employee recounted, expressing disbelief. Yet, GSA leadership abruptly reversed course. Shortly before the planned announcement of partnerships with OpenAI, Anthropic, Google’s Gemini, and xAI, staff were instructed to remove xAI’s Grok from the contract offering. Two GSA workers involved in the contract believe Grok’s controversial outputs were the direct cause of xAI’s exclusion.
The rapid pace of these partnerships also presented practical challenges. For instance, the agreements with OpenAI and Anthropic were finalized so swiftly that “it wasn’t even clear who to send the $1 to or how,” according to one GSA source. Furthermore, while OpenAI and Anthropic have developed tools specifically for government use, neither company has fully cleared the necessary regulatory hurdles, such as FedRAMP approval—a GSA-led program ensuring the security of private cloud services through rigorous screenings. However, implementation memos include carve-outs allowing non-FedRAMP-approved products to be used in a research and development capacity. A former White House official criticized the administration for issuing an executive order that mandated such a rapid turnaround, which precluded consultation with key stakeholders.
The Trump administration’s aggressive pursuit of AI integration is evident across various agencies. At the Centers for Medicare and Medicaid Services, for example, there have been suggestions of replacing frontline health workers with AI avatars. Representatives from the Department of Government Efficiency (DOGE) have utilized AI to identify regulations for elimination and to write code. In June, US spy chief Tulsi Gabbard spoke about employing AI tools to review classified documents related to the assassination of John F. Kennedy, though these files were later found to contain sensitive personal information of living individuals. An April memo at the Department of Veterans’ Affairs instructed staff to compile rules and regulations for AI review, aiming to narrow the field of documents requiring further human scrutiny. A June draft memo from the VA projected that “within the next 1 to 3 years, most computer-based tasks at VA will be automatable,” including AI-powered digital assistants assisting veterans with benefit and healthcare transactions. The GSA itself has launched its own government chatbot, GSAi, encouraging its integration into daily workflows, with plans to link it to the agency’s internal data sources for real-time information.