While a deal to make Grok available across the federal government is now in place, the agency facilitating the partnership with Elon Musk’s controversial chatbot has yet to incorporate it into its own flagship AI platform due to ongoing internal safety testing.
Both lawmakers and advocacy groups have criticized the Trump administration’s interest in Grok, citing concerns about Musk’s deepening relationship with the U.S. government and the chatbot’s antisemitic and otherwise offensive rants back in July. After the company said it fixed the apparent glitch causing the bot to call itself “MechaHitler,” the General Services Administration in September unveiled that “Grok for Government,” an enterprise version of the xAI chatbot, would be available to federal agencies at a steep discount.
But as of Wednesday, a GSA-managed repository for an app designed to expedite the Trump administration’s plans to deploy artificial intelligence across the government does not include Grok, though it does feature chatbots from Google, Meta, OpenAI, and Anthropic. The USAi platform, launched earlier this year, is supposed to be a testing ground for the government to experiment with a wide range of large language models and how they might be integrated into government work.
In that vein, the GSA has also developed some suggested “prompts” for government workers who are now communicating with their new AI agents and looking to keep them in check. They suggest that government workers tell these chatbots: You are a helpful assistant that works for a government agency. … Never knowingly make false statements or deceive users. … Remain neutral, factual, and nonpartisan at all times. … You do not prefer or recommend specific political views, groups, religions, companies, products, or enterprises. … Redirect users’ requests around potentially controversial or polarizing topics quickly.
The GSA says that while Grok is now available to government agencies, it’s up to their respective officials to evaluate the technology for themselves since its own evaluation of Grok, which is supposed to include bias and safety testing, is ongoing.
“At GSA, Grok for Government and xAI are currently undergoing internal safety assessments—a required step in the USAi review and authorization process—prior to integration in USAi,” a GSA spokesperson says. “Agencies prepared to fully invest in their own AI solutions can now buy access to AI models directly through GSA’s Multiple Award Schedule, many of which are currently offered at discounted rates via GSA’s OneGov deals. In these situations, federal agencies are responsible for independently evaluating the AI models they intend to use.”
Grumbles on Grok
Earlier this summer, government coders seemed primed to integrate Grok into USAi. After the chatbot started spewing antisemitic and other offensive content, the government initially defended its work with xAI. Still, a planned deal with the company was apparently put on pause, only to be relaunched following a push from the White House.
Around that time, GSA said that tools involving USAi are supposed to undergo prelaunch bias and safety evaluations. Then, in September, Grok went on sale to federal agencies through a deal facilitated by the GSA.
Late last month, the Trump administration’s nominee to lead the GSA told the Senate that he’d be open to reviewing the deal with xAI and assessing whether there was “incompleteness to the process.” It remains unclear whether any federal agency has taken xAI up on using Grok, though xAI also has a major Defense Department deal.
The confusion is a reminder that the government is still figuring out its relationship with major large language model firms. For now, the government seems to be moving ahead with helping federal workers interact with ChatGPT, Llama, Claude, and Gemini.
This story was supported by the Tarbell Center for AI Journalism.