Anthropic and OpenAI Say China's Open AI Is Dangerous

Anthropic and OpenAI are lobbying Washington to restrict Chinese open-source AI models, framing it as a safety issue. Nvidia's Jensen Huang, Microsoft's Satya Nadella, Elon Musk, Mark Zuckerberg, Sam Altman himself on other fronts, and eventually Google's Sundar Pichai all publicly pushed back within 72 hours, per reporting from the New York Times. When nearly the entire industry, including some of the people the safety argument is supposed to protect, lines up against you within three days, the safety framing deserves a second look.

Key Points

• Anthropic and OpenAI have lobbied regulators, including Treasury Secretary Scott Bessent and White House science adviser Michael Kratsios, arguing Chinese open-weight models are too risky to leave unrestricted
• Nearly 200 startups under the "Little Tech Association" signed letters opposing restrictions on Chinese open-source models the same week
• Jensen Huang and Satya Nadella posted public statements nine minutes apart backing open source, followed by a joint letter also signed by Meta, Palantir, and IBM
• Chinese startups Z.ai and Moonshot AI released models in June and July that reportedly rival American labs' output, which is the timeline that triggered the lobbying push
• Anthropic and OpenAI have also claimed China's models were built by improperly harvesting data from their own systems, a claim that remains unproven

Follow Where Each Company Makes Money

Anthropic and OpenAI sell closed, proprietary access to frontier models. That's the entire business. A cheap, capable, freely available Chinese alternative is a direct threat to that model, not a bomb-making manual. Nvidia and Microsoft sell chips and cloud compute; open-source models drive demand for both regardless of who built them, which is exactly why Huang and Nadella don't share the same incentive to restrict access. Meta, which has its own open-weight models, obviously doesn't either. The safety argument and the balance-sheet argument point in opposite directions depending on which company is making it, and that's worth sitting with before accepting either one uncritically.

The Timing Gives the Game Away

This fight didn't escalate because Chinese models suddenly became more dangerous. It escalated because Z.ai and Moonshot AI released models in June and July that reportedly rival American labs on performance, for free. Anthropic and OpenAI didn't start lobbying Washington when the safety concern first existed, if it ever did in the form they're describing. They started when the competitive threat became undeniable. That sequencing is the whole story.

"Too Dangerous to Be Open" Is Doing a Lot of Work Here

The framing itself deserves scrutiny. Huang's position, stated plainly on X, is that the world needs both frontier closed models and frontier open models. That's a defensible technical position independent of any China angle: open-weight models let more people examine and stress-test the security of a system, which is the opposite of what Anthropic and OpenAI are implying. If openness were the actual risk, the argument would need to explain why Meta's open models get a pass while China's don't. It mostly doesn't try.

What This Means If You're Building on These Tools

Marketers and growth teams evaluating AI vendors should treat "safety" claims from any lab, American or Chinese, as a business position until proven otherwise, not a settled fact. The practical question isn't which flag a model was built under. It's whether the restriction being proposed protects you or protects a competitor's market position. If you're weighing model choice as part of a broader AI marketing services decision, that distinction should shape the conversation more than either side's press strategy.

This is a market-share fight wearing a national-security costume, and most of the room seems to know it.