I’m confused by Anthropic’s latest blog post:

Is this really about national security, or is it also about corporate self-interest?

• A lot of models coming out of Chinese labs are open-source or released with open weights (DeepSeek-R1, Qwen series), which has clearly accelerated accessibility and democratization of AI. That makes me wonder if Anthropic’s move is less about “safety” and more about limiting potential competitors.
• On OpenRouter’s leaderboard, Qwen and DeepSeek are climbing fast, and I’ve seen posts about people experimenting with proxy layers to indirectly call third-party models from within Claude Code. Could this policy be a way for Anthropic to justify blocking that kind of access—protecting its market share and pricing power, especially in coding assistants?

Given Dario Amodei’s past comments on export controls and national security, and Anthropic’s recent consumer terms update (“users must now choose whether to allow training on their data; if they opt in, data may be retained for up to five years”), I can’t help but feel the company is drifting from its founding ethos. Under the banner of “safety and compliance,” it looks like they’re moving toward a more rigid and closed path.

Curious what others here think: do you see this primarily as a national security measure, or a competitive/economic strategy?

full post and pics: https://x.com/LuozhuZhang/status/1963884496966889669