The White House did not immediately respond to a request for comment.
One company that was supposed to be part of the GSA announcements was Musk’s xAI, according to sources with knowledge of the discussions.
In early June, GSA leadership met with the xAI team for a two-hour brainstorming session “to see what opportunities may exist for automation and streamlining,” according to an email obtained by WIRED.
The session appeared to go well. After it ended, GSA leadership continued to push for the agency to roll out Grok for internal use. “We kept saying ‘Are you sure?’ And they were like ‘No we gotta have Grok,’” one employee involved with the discussions tells WIRED.
The conversations went far enough that xAI was added to the GSA Multiple Award Schedule, which is the agency’s long-term government-wide contracting program, by the end of June, according to documents obtained by WIRED. The move would have allowed federal agencies to buy Grok through Carahsoft, which is a technology reseller and government contractor.
Then, in early July, Grok appeared to go off the rails, spewing antisemitic hate, praising Adolf Hitler, and parroting racist conspiracy theories on X. Some GSA staffers were surprised that the incident did not appear to slow down the procurement process. “The week after Grok went MechaHitler, [GSA leadership] was like ‘Where are we on Grok?’” the same employee claims. “We were like, ‘Do you not read a newspaper?’”
Then GSA leadership appeared to abruptly change course. Shortly before GSA was set to announce its partnerships with OpenAI, Anthropic, Google’s Gemini, and xAI last week, staff were instructed to remove xAI’s Grok from the contract offering, two sources with knowledge tell WIRED. Two GSA workers involved with the contract believe xAI was pulled because of Grok’s antisemitic tirade last month.
xAI did not immediately respond to a request for comment from WIRED.
The announcements for GSA’s partnerships with OpenAI and Anthropic, meanwhile, happened so quickly that “it wasn’t even clear who to send the $1 to or how,” one GSA source tells WIRED.
And while OpenAI and Anthropic have released tools exclusively for government use, neither company has cleared the regulatory hurdles necessary for agencies to buy directly from them. Specifically, they haven’t been approved through FedRAMP, a GSA-led program that ensures the safety of private cloud services through intense security screenings. There are, however, carve-outs within the implementation memo to allow for products that have not been FedRamp-approved to be brought into government in a research and development capacity.
“It was irresponsible of the administration to issue an executive order that required such a fast turnaround to get those implementation memos out,” because it meant that the government was unable to consult with the significant number of stakeholders that they would otherwise have, says a former White House official who spoke to WIRED on the condition of anonymity.
The Trump administration has wasted no time bringing AI into government. One of the first executive orders signed by Trump prompted agencies to reverse any rules inhibiting the growth and dominance of American AI, kicking off a mad dash among administration leaders to find new ways to incorporate the tech into everything. At the Centers for Medicare and Medicaid Services, Mehmet Oz has suggested replacing some frontline health workers with AI avatars. Representatives of the so-called Department of Government Efficiency used AI to find regulations to slash and write code. In June, US spy chief Tulsi Gabbard spoke at an Amazon Web Services summit about having used AI tools to review classified documents related to the assassination of John F. Kennedy. (When released, the files turned out to contain the Social Security numbers and additional private information of hundreds of living people.)