OpenAI, Google & Perplexity Near US Government AI Contracts | Exclusive
- OpenAI, Google, and Perplexity AI are nearing approval to sell their artificial intelligence technologies directly to the U.S.
- Currently, many AI firms rely on established government technology contractors – including Microsoft, Palantir, and Amazon Web Services – to host their AI models for federal users.
- The drawbacks of this dependence are currently playing out in a public dispute between the Pentagon and Anthropic, an AI safety-focused company.
AI Companies Move Towards Direct Government Sales, Circumventing Traditional Contractors
OpenAI, Google, and Perplexity AI are nearing approval to sell their artificial intelligence technologies directly to the U.S. Government, a significant shift that could reshape how federal agencies procure and deploy AI solutions. The authorization, initially expected at a “low impact” and pilot level, represents a move towards greater independence for these AI companies and a potential streamlining of the government’s AI acquisition process.
Currently, many AI firms rely on established government technology contractors – including Microsoft, Palantir, and Amazon Web Services – to host their AI models for federal users. This arrangement, while providing a faster route to market by leveraging the contractors’ existing security clearances and government relationships, has also meant ceding some control over access and application of their technologies. The reliance on intermediaries adds complexity and cost to the procurement process.
Pentagon-Anthropic Dispute Highlights Risks of Dependence
The drawbacks of this dependence are currently playing out in a public dispute between the Pentagon and Anthropic, an AI safety-focused company. The Defense Department is threatening to terminate a $200 million contract with Anthropic after the AI firm requested limitations on how its Claude model is used, specifically prohibiting applications in autonomous weapons systems and mass surveillance. This conflict was reportedly exacerbated by Anthropic’s partnership with Palantir.
According to reports, Anthropic’s Claude model was utilized in a U.S. Operation to capture former Venezuelan President Nicolás Maduro, facilitated through Palantir’s systems. Tensions escalated when an Anthropic official inquired with a Palantir executive about the model’s specific use in the operation, raising concerns within the Pentagon about Anthropic’s commitment to supporting military applications. This situation underscores the potential for misalignment between an AI developer’s ethical guidelines and a contractor’s operational requirements when a third party controls deployment.
FedRAMP 20x Expedites Security Reviews
To achieve greater independence, OpenAI, Perplexity, and Google proactively applied for and received expedited review of their cloud systems last year under FedRAMP 20x, a federal security initiative designed to accelerate the authorization of cloud services. Approval under FedRAMP 20x signifies that these companies have taken substantial steps to meet the government’s stringent security requirements and are now poised to engage directly with federal agencies. These approvals are distinct from individual agency purchasing decisions but demonstrate a commitment to direct engagement.
In contrast, Anthropic has heavily relied on partners like Palantir to facilitate sales of its technology to government customers. The company does not appear to have participated in the FedRAMP 20x program, though the reasons for this are unclear. However, Anthropic has publicly acknowledged the benefits of direct access, with Michael Sellito, the company’s head of global affairs, stating in that the company “would also like to be able to directly provide services to governments and not necessarily go through a partner at all times.”
Perplexity AI Secures First Direct-to-Agency Deal
This shift towards direct engagement is already manifesting in concrete deals. The General Services Administration (GSA) recently struck an agreement with Perplexity AI to offer its artificial intelligence services to all federal agencies for 25 cents per use, marking the first contract under the GSA’s OneGov initiative. This deal represents a significant departure from traditional procurement methods, offering a direct-to-consumer product for federal workers and bypassing the need for intermediary agencies that typically resell contracts at a higher cost.
Perplexity’s AI services will provide federal employees with access to real-time information, combining internet-scale knowledge with connections to the government’s existing technological infrastructure. Josh Gruenbaum, commissioner of GSA’s Federal Acquisition Service, emphasized that this type of deal is “essential for integrating AI into federal government.” Jerry Ma, Perplexity’s vice president of policy, added that as the federal government increases its use of AI, its needs from leading-edge AI providers will only grow, positioning Perplexity as a “natural partner” for that evolution.
Broader Implications for AI Development and Adoption
GSA officials believe that deals like the one with Perplexity will foster AI growth within the U.S. By expanding the use cases for AI systems and accelerating their development, while simultaneously enhancing the capabilities of federal workers. The move towards direct engagement is expected to encourage further innovation and competition within the AI sector, potentially leading to more efficient and effective government services. The GSA’s OneGov initiative aims to modernize government procurement, and the Perplexity deal serves as a model for future direct-to-agency contracts.