American military planners face an uncomfortable reality: the artificial intelligence systems being pitched by Silicon Valley's most prominent firms are increasingly at odds with the operational security and sovereignty required for national defense. Retired Major General James 'Spider' Marks is now publicly cautioning that closed, proprietary AI models introduce a strategic vulnerability—one that adversaries are actively working to exploit.
The core of the concern rests on control. Closed commercial models, such as those fielded by Anthropic and OpenAI, operate behind impenetrable proprietary architectures. Their corporate owners can modify algorithms, alter safety guardrails, or revoke access overnight. A recent example cited involves Anthropic's Fable 5 model, which researchers reported refused to answer benign educational questions and selectively degraded responses related to advanced AI development, effectively shaping access to knowledge according to a private company's interests. For a military commander, that level of external dependency is unacceptable in a weapon system, communications network, or intelligence platform.
"No military commander would accept single point-of-failure dependence in weapon systems. Open models offer genuine operational control—they can be deployed on sovereign infrastructure, audited for security, and tailored to mission requirements independently of any commercial provider's priorities."
The strategic calculus is complicated by the narrowing performance gap with Chinese AI models. Beijing-based startup Moonshot AI recently unveiled its Kimi K3 model, demonstrating front-end coding capabilities comparable to American titans. At the 2026 World Artificial Intelligence Conference in Shanghai, President Xi Jinping positioned China as a champion of open-source artificial intelligence. This approach, according to a recent JPMorganChase report, gives China distinct strengths in the AI race despite America's leadership in semiconductor design and financing.
The U.S. defense establishment is already responding by pursuing a "best-of-breed" strategy, partnering with multiple frontier companies—including providers of transparent, open-weight models. This diffused approach marks a departure from rigid dependence on a small coterie of defense contractors and seeks to prevent any single firm from leveraging regulatory capture to lock in its market position. The path to maintaining American primacy in AI, according to military thinkers, requires cultivating a broad domestic innovation base free from the gatekeeping of a few Silicon Valley boardrooms whose interests are not necessarily aligned with the nation's own.