The White House did not answer immediately at the request for comment.
One of the companies that was to be part of GSA announcements was Xai Muska, according to sources with knowledge about discussions.
At the beginning of June, GSA Leadership met with the XAI team during a two-hour brainstorming session “to see what possibilities can exist for automation and improvement”, according to the e-mail obtained by Wired.
The session seemed good. After the end of GSA, the leadership continued to press the agency to bring the groc into internal exploit. “We still said:” Are you sure? “And they were like” no, we must have groc “, one employee involved in the discussion says Wired.
The talks have gone far enough for XAI to be added to the multiple GSA award schedule, which is a long -term program arranged by the agency, by the end of June, in accordance with the documents obtained by Wired. This movement would allow federal agencies to buy groc by Carahsoft, who is a technology seller and government contractor.
Then, at the beginning of July, the grok seemed to get off the rails, throwing anti -Semitic hatred, praising Adolf Hitler and racist parrot conspiracy theories about X. Some GSA employees were surprised that the incident did not slow down the order process. “Mechahitler went a week after Groza, [GSA leadership] It was like “where are we at the grok?” The same employee claims. “We were like:” Don’t you read the newspaper? “
Then the GSA leadership seemed to suddenly change the course. Shortly before GSA, it announced its partnerships with OpenAI, Anthropic, Google’s Gemini and XAI last week, employees were instructed to remove Grook XAI from the offer of the contract, two sources with Knowledge Tell Wired. Two GSA employees involved in the contract believe that XAI was pulled out of the anti -Semitic Tyrada Grok last month.
Xai did not immediately answer at the request for comment from Wired.
Meanwhile, advertisements about the GSA partnership from OpenAI and Anthropic took place so quickly that “it was not even clear to whom to send $ 1 or how,” says GSA Source.
While Opeli and Anthropic published tools adapted to government use, no company has achieved the independent Fedramp authorization, allowing them to directly sell all its products. Fedramp is a program run by GSA, which ensures the security of private services in the cloud through intensive security research. In the implementation note, however, there are reproach to enable products that have not been approved Fedramp, which will be introduced into the government in the field of research and development.
“The irresponsible administration was issuing an executive ordinance that required such a quick return to issue these implementation notes,” because it meant that the government was unable to consult a significant number of stakeholders that they would have differently, says a former White House official who talked to faith in the conditions of anonymity.
The Trump administration did not waste time to introduce AI to the government. One of the first executive orders signed by Trump prompted the agencies to reverse all principles inhibiting the growth and dominance of the American artificial intelligence, starting a crazy board among administration leaders to find new ways to include technology in everything. At the Center for Medicare and Medicaid Services, Mehmet Oz suggested replacing some of the AI Avatars first -line healthcare professionals. Representatives of the so -called government department used artificial intelligence to find recipes for cutting and writing the code. In June spoke At the Amazon Web Services top on the exploit of AI tools for reviewing classified documents related to the murder of John F. Kennedy. (After spending, the files turned out to be included Numbers of social insurance and additional private information of hundreds of living people.)