The U.S. Department of Defense (DoD) has significantly reduced its reliance on AI company Anthropic, shifting the majority of its AI workloads to other major tech firms. This move operationalizes a plan to diversify its AI suppliers, driven by concerns over security, procurement, and dependency on a single vendor.
The chain of events leading to this decision began in earnest earlier this year. First, in February 2026, a critical disagreement arose when Anthropic refused a DoD request to remove certain safety 'guardrails' from its AI models, particularly concerning autonomous weapons and surveillance. This refusal was a pivotal moment, setting the stage for a major policy shift.
Second, following this impasse, the DoD took the decisive step in March 2026 of labeling Anthropic a 'supply-chain risk'. This is a serious designation typically reserved for foreign adversaries, and it immediately triggered an urgent need to find and implement alternatives to Anthropic's AI model, Claude, which had become widely embedded in DoD systems.
Third, to facilitate this transition, the Pentagon acted swiftly. By May 2026, it had signed agreements with seven other AI companies, including giants like OpenAI (via Microsoft), Google, and NVIDIA. These deals provided the necessary infrastructure and alternative models to begin migrating workloads off of Anthropic's platform, effectively creating a multi-vendor environment.
This diversification strategy also aligns with the DoD's broader budgetary priorities. Recent budget requests, totaling over $50 billion for AI and autonomy, emphasize building a 'sovereign AI infrastructure'. Such an architecture is designed to prevent 'vendor lock-in'—a situation where an organization becomes overly dependent on a single supplier—and ensure the U.S. military can scale its AI capabilities across various services and classification levels without disruption.
- Supply-chain risk: A designation given to a supplier or company that is deemed to pose a threat to the security and integrity of an organization's operations or data.
- Guardrails: In AI, these are safety rules and filters designed to prevent the model from generating harmful, unethical, or dangerous outputs.
- Vendor lock-in: A situation where a customer using a product or service cannot easily transition to a competitor without substantial switching costs or operational disruption.