OpenAI recently signed a public letter backing open-weight artificial intelligence, while Anthropic chose to sit it out. Behind closed doors in Washington, both are lobbying lawmakers to restrict open-weight models from China, citing national security risks. What they say publicly and what they say to policymakers are two different things.
The wider coalition that signed the letter, including Nvidia, Microsoft, Meta and Hugging Face, made the case against premature restrictions. The argument is simple: restricting open-weight models narrows participation, concentrates power in a small number of frontier labs and makes AI governance harder for everyone outside that group. That’s OpenAI’s public position, though seemingly not its private one.
The Fine Line Between National Security And Market Protection
Lawmakers in Washington have trained their attention specifically on Chinese open-weight artificial intelligence. The stated rationale is that accessible Chinese systems endanger national defence, justifying tight restrictions on their use across American infrastructure. Independent researchers from diverse political backgrounds validate that security risk. The complication is that “restrict Chinese open-weight models” and “restrict open-weight models broadly” aren’t the same policy, and the second tends to follow from the first.
OpenAI and Anthropic have strong commercial reasons to want the open-weight environment to be less permissive. Their core business model depends on users paying for API access to frontier models. A world where high-quality open-weight models are freely available and widely used is a world where the justification for that pricing is harder to maintain. The national security argument holds genuine merit while also advancing the commercial interests of the firms championing it.
More from Artificial Intelligence
- Andy Burnham Launches New AI Taskforce To Drive Economic Growth
- Meta Has Pledged To Be Net Zero By 2030. So Why Is It Walking Away From RE100?
- New Data Reveals How Much AI Is Costing UK Business Operations
- SaaS Founders Are Panicking About AI But A $21 Billion Bet Says They’re Wrong
- Anthropic Launches Claude Certifications To Help Businesses Choose AI Partners
- Are Companies Still Laying Off Employees For AI?
- AI Minister In, Tech Department Out – Is The UK Sidelining Tech For AI?
- The Apple Store Wants AI To Help You Make Purchases – How Will It Work?
China’s Countermove
Xi Jinping took the stage at WAIC 2026 to promise 5,000 artificial intelligence training placements, joint regional centres and tool access for developing economies. This wasn’t rhetorical positioning. China is bundling open software directly into its foreign aid strategy, positioning its software as a sensible discount alternative to expensive American software licensing.
The geopolitical outcome is easy to predict. If US policy narrows access to open-weight models, particularly Chinese ones, and US frontier labs remain expensive for cost-sensitive markets, the vacuum gets filled. China is moving to fill it with open access, technical cooperation and the implicit message that its AI infrastructure comes without the geopolitical strings attached to US alternatives. That’s a stronger offer to a finance ministry in Lagos or a university in Jakarta than a $20-per-million-token API rate.
Chinese AI models, including DeepSeek and Qwen, are already a lot cheaper than comparable US models. In several benchmarks they’re competitive on capability. The cost difference alone makes them attractive in markets where budget constraints dominate procurement decisions. Open-weight availability makes them usable without ongoing API dependency.
When Neutral Software Architecture Turns Political
Founders who built product assumptions around open AI being stable, neutral and globally accessible are now sitting in the middle of a policy fight they didn’t cause and can’t control. The promise of open-weight models was that access would be technically determined: good models, free to run and no API dependency. That premise is now under pressure.
If US policy hardens against Chinese models, implementing or building on Qwen or DeepSeek becomes complicated in ways that weren’t previously the case. If US frontier labs reduce pressure on open-weight availability through lobbying, the free alternative to expensive US APIs shrinks too. The middle ground, truly open and neutral AI, is what independent builders need most. It’s also what both Washington lobbying and Chinese geopolitical strategy are working to eliminate.
The real battle isn’t open source versus closed source. It’s whether openness becomes a strategic export tool for China while US policy makers try to regulate it as a national security risk. The premise that model access, licensing and compute would depend strictly on technical merit instead of political strategy has proven unreliable for founders. Model infrastructure is part of the product risk stack in a way it wasn’t two years ago.
