The US Department of Defense has migrated approximately 90% of its classified AI workloads away from Anthropic's closed-source models. The agency expects to complete this transition by the end of September 2026. Earlier in 2026, the Pentagon designated Anthropic as a national security supply-chain risk due to the company's refusal to provide unrestricted access to its models for government use.
The conflict stems from Anthropic's usage policies, which restrict applications like mass surveillance and autonomous lethal systems. Defense agencies require full control and auditability of AI tools. Booz Allen Hamilton's Cyber Weapon Index, released in early September 2026, highlighted that Anthropic's Claude Mythos model could autonomously execute cyber kill chains, further complicating its use in sensitive environments.
Nvidia and Palantir have formed a partnership, announced on September 10, 2026, to provide sovereign AI solutions. Nvidia offers its Nemotron models, designed for controlled deployments, while Palantir integrates these tools into military logistics. Additionally, Nvidia launched the Open Secure AI Alliance in July 2026 to promote open-weight models for national security applications.
The shift toward open-weight alternatives is supported by major defense contractors and agencies seeking to avoid dependencies on private ethical frameworks. Meta's Llama models are also positioned to benefit from this trend, as the defense sector increasingly prioritizes the ability to inspect and control AI systems over the safety-first restrictions imposed by providers like Anthropic.
