Eight leading technology companies have secured approval to deploy their artificial intelligence systems on the Pentagon's classified computer networks, marking a significant expansion of AI capabilities within the most sensitive military and intelligence infrastructure. The agreements cover networks handling secret data through top-level classified systems, though notably exclude Anthropic despite its existing use in military applications.
Eight Companies Gain Access to Secret-Level Networks
The approved companies—Amazon Web Services, Google, Microsoft, OpenAI, SpaceX, NVIDIA, Reflection, and Oracle—can now deploy their AI systems on networks classified as Impact Level 6 (handling secret data) and Impact Level 7 (the most highly classified systems). This represents the first comprehensive framework allowing commercial AI models to operate within the Pentagon's most sensitive computing environments.
The inclusion of OpenAI is particularly significant given previous debates about military applications of their technology. SpaceX's approval suggests expansion beyond traditional defense contractors into companies with dual-use capabilities.
Government Pre-Deployment Testing Framework Established
Beyond Pentagon deployment, Microsoft, Google, and xAI have agreed to allow the United States federal government access to their new AI models for national security testing. The Center for AI Standards and Innovation (CAISI) at the Department of Commerce announced this separate framework, which enables government evaluation before public deployment.
Under these agreements, developers provide versions of their models with safety guardrails stripped back, allowing CAISI to probe for national security risks including dual-use capabilities, adversarial vulnerabilities, and potential misuse scenarios.
Anthropic's Notable Exclusion Despite Existing Military Use
Anthropic's absence from the approved list is particularly striking given that its Claude AI already operates on classified networks as part of Palantir's Maven toolkit. According to reports, the administration has attempted to ban Anthropic from government work, triggering multiple lawsuits. This exclusion suggests policy-driven decisions rather than purely technical evaluations, as Claude has demonstrated functional capability in classified environments through the Palantir integration.
The Maven toolkit, used for intelligence analysis and military applications, has successfully integrated Claude for processing classified information, making the broader Pentagon exclusion a departure from operational reality.
Key Takeaways
- Eight companies (AWS, Google, Microsoft, OpenAI, SpaceX, NVIDIA, Reflection, Oracle) approved for Pentagon classified networks at Impact Levels 6 and 7
- Separate pre-deployment testing agreements established with Microsoft, Google, and xAI through the Department of Commerce's CAISI
- Anthropic excluded from Pentagon agreements despite Claude's existing use in Palantir's classified Maven toolkit
- Government testing framework requires developers to provide models with safety guardrails removed for national security evaluation
- Agreements create two-tier framework: deployment authorization for classified networks and pre-deployment security assessment