The Pentagon announced on 1 May 2026 that eight technology companies are now cleared to deploy their AI systems on the Department of Defense's most sensitive classified networks. The approved firms are Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, SpaceX, Reflection (an NVIDIA-backed startup), and Oracle, which was added later the same day. The clearance covers Impact Level 6 networks handling secret data and Impact Level 7 systems for the most highly classified information. The stated objective is to 'streamline data synthesis, elevate situational understanding, and augment warfighter decision-making in complex operational environments.'
The most conspicuous absence is Anthropic. Despite Claude already being used on classified networks via Palantir's Maven toolkit, the company was excluded from the new agreements. The Trump administration had previously attempted to ban Anthropic from government work entirely after negotiations over safety guardrails broke down in February, leading to lawsuits. Pentagon CTO Emil Michael stated that 'it is irresponsible to be reliant on any one partner' and referenced a partner that 'did not really want to work with us' — widely interpreted as a reference to Anthropic's insistence on contractual safety limits around surveillance and autonomous weapons use.
For context engineers, the geopolitical dimension of AI development is now impossible to ignore. OpenAI's path to classified military deployment began in March when it signed a deal with the Pentagon — which CEO Sam Altman admitted was 'rushed' and 'looked opportunistic and sloppy' — after Anthropic's negotiations collapsed. The deal includes red lines prohibiting mass domestic surveillance, autonomous weapons systems, and high-stakes automated decisions, though critics including the Electronic Frontier Foundation have called these 'weasel words.' The broader picture is that frontier AI models are now being deployed in military contexts at a pace that has outrun the policy frameworks meant to govern them. For developers building on these platforms, the question of which company's values and safety commitments you trust is no longer abstract — it directly shapes what your tools are being used for.
