American closed models vs Chinese open models is becoming a problem.

A security professional working with government and defense clients has exposed a critical vulnerability in the US AI stack: a severe shortage of capable, open-weight models for secure, offline deployment. While Chinese firms like Zhipu AI (GLM series) and MiniMax are rapidly advancing and releasing powerful open models, the US ecosystem is dominated by closed, API-only offerings from leaders like OpenAI and Anthropic. The only notable recent US open model, GPT-OSS-120B, is considered far behind the curve, forcing sensitive industries into an impossible trade-off between national security mandates and technological capability. This gap is reportedly creating pressure on companies like Anthropic from entities like the Department of Defense, which need cutting-edge AI for secure applications.

The technical bind is stark. For environments where data cannot leave a closed network, open weights are essential. The professional notes that covertly using superior Chinese models is not an option, leaving few alternatives: lobbying US AI giants for open releases, hoping for international options from companies like Canada's Cohere, or falling behind. This situation underscores a strategic risk where US national security policy may inadvertently cede technological leadership in a crucial sector. The post suggests the resolution may require either a policy shift or a significant new commitment from US AI labs to open-source development to ensure sovereign capability.