OpenAI, Google, and Perplexity near approval to host AI directly for the U.S. government (exclusive)

Three AI companies—OpenAI, Google, and Perplexity—are on the verge of receiving approval to sell their technology, hosted on their own cloud systems, directly to the government, a person familiar with the matter tells Fast Company. That authorization will be on a “low impact” and pilot level, the person said, but constitutes a major step toward independence.

That independence could help those companies avoid some of the complications created by ongoing partnerships between AI firms and longtime government tech contractors. As large language models have gone mainstream, AI companies have often relied on tech firms that have already passed arduous government security reviews—including Microsoft, Palantir, and Amazon Web Services—to host their chatbots for federal users. In the early days, these partnerships made it easier for AI labs to quickly get their tech in front of government officials, but also meant ceding at least some control over when and how their AI was made available.

The downside of that kind of dependence is now playing out in the brewing feud between Anthropic and the Pentagon, which appears to have been fueled, in part, by its partnership with Palantir. The Defense Department is threatening to cancel a $200 million contract after Anthropic requested limits on the use of its AI for certain applications, including autonomous weapons and mass surveillance.

Anthropic’s Claude model was made available to military officials with the help of Palantir’s systems and was even used in the U.S. operation to capture former Venezuelan President Nicolás Maduro, according to reports. According to Semafor, tensions mounted after an Anthropic official asked a Palantir executive how Claude had been used in the operation, prompting concern inside the Pentagon about the company’s willingness to support military applications.

This is all to say that not relying on a company like Palantir makes selling to the government far less complicated. In pursuit of that independence, OpenAI, Perplexity, and Google applied for, and received, expedited review of their cloud systems last year under a federal security initiative called FedRAMP 20x. Now, Fast Company can report, they’re almost certain to be approved.  These approvals are separate from any decision by a specific federal agency to purchase their products, but they show the companies have taken concrete steps to engage the government on their own terms.

Anthropic, by contrast, has leaned heavily on partners like Palantir to help sell its technology to government customers. The company does not appear to have participated in FedRAMP 20x, though it’s not clear why. Still, the question of independence is one Anthropic has publicly acknowledged. “We would also like to be able to directly provide services to governments and not necessarily go through a partner at all times,” Michael Sellito, the company’s head of global affairs, told FedScoop in 2024.

Neither Palantir nor Anthropic responded to Fast Company’s request for comment.

View the full article