The U.S. Department of Defense has selected 7 companies to participate in classified AI procurement... including OpenAI and Google, with Anthropic excluded.

The U.S. Department of Defense announced the list of seven companies awarded AI procurement contracts. Including Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, SpaceX, and the relatively lesser-known startup Reflection AI, while Anthropic was excluded.

This contract aims to expand the scope of services provided through the Department of Defense’s internal AI portal “GenAI.mil.” According to foreign media reports, over 1.3 million Department of Defense personnel have used the platform since its launch last year, building hundreds of thousands of AI agents. The Department is focusing on this to reduce data integration work and accelerate decision-making processes.

Highest-level environment for handling classified information

The AI products involved in the new contracts will be applied in “Impact Level 6” and “Impact Level 7” environments. These are system levels where the Department of Defense can store classified information, effectively corresponding to the highest security requirements. Although specific products are not publicly disclosed, this procurement clearly has moved beyond mere experimentation into “classified AI operations.”

The assets that companies may provide are also of great interest. While NVIDIA is known for graphics processing units, it also has neural network development tools and open-source large language models. Particularly, its models based on the “Mamba-Transformer” architecture, which reduces memory usage, are believed to have potential applications in limited computing environments such as defense.

SpaceX and Reflection AI are also under attention… Anthropic excluded

SpaceX’s inclusion is also noteworthy. This year, SpaceX merged with xAI Holdings and acquired the “Grok” series of language models, strengthening its presence as an AI provider. Additionally, its potential expansion into programming-specific model portfolios has led some to believe it could accelerate its transformation from an aerospace company into a defense AI supplier.

Reflection AI is the most unfamiliar name on this list. Founded in 2024 by researchers from Google DeepMind, this startup has not yet launched commercial products but raised $2 billion last year, approximately 2.95 trillion Korean won. U.S. media reported that the company is preparing to launch a language model trained on hundreds of trillions of tokens.

In contrast, the exclusion of Anthropic was almost expected. Defense Secretary Pete Hegseth designated Anthropic as a “supply chain risk” target in February this year, and the measure was formalized in March. Accordingly, the Department of Defense is prohibited from using Claude, and access for defense contractors is restricted.

Background of banning Claude — conflict with “all lawful purposes” clause

The core conflict lies in the contract terms. Secretary Hegseth stated that the sanctions were implemented because Anthropic does not accept the Department of Defense’s requirement for “use for all lawful purposes.” Anthropic worries that such language could leave room for AI models to be used for domestic mass surveillance or autonomous weapons development. Ultimately, the company filed a lawsuit in March this year against the supply chain risk designation.

However, the situation is not so simple. According to foreign media reports, despite the ban, the Department of Defense is reportedly still using Anthropic’s “Claude Mitos Preview.” Although adopted by some other federal agencies, it has not been publicly released. The reason is its exceptional ability to detect previously unidentified security “zero-day” vulnerabilities.

This AI procurement contract from the U.S. Department of Defense is not just about technology acquisition but is also seen as a signal indicating which companies’ AI systems can be trusted within the national security framework. At the same time, Anthropic’s case shows that performance alone is not enough to enter the defense market; contract terms and policy acceptance have become core variables in the AI supply chain.

TP AI notes: The article was summarized using TokenPost.ai’s basic language model. Some main content may have been omitted or may not align with facts.