OpenAI and Anthropic Lobby Against Chinese Open-Weight AI Models

U.S. AI Labs Align to Restrict Chinese Open-Weight Models

OpenAI and Anthropic are coordinating efforts to warn U.S. policymakers about the risks posed by powerful open-weight AI models from China. While these two companies compete fiercely for market share, they have found common ground in advocating for greater scrutiny and potential restrictions on models where the weights are publicly available, arguing that such openness undermines safety and intellectual property.

The Safety vs. Control Debate

Closed-model labs argue that open-weight models are inherently less safe because the developers cannot revoke access, update safety guardrails, or prevent misuse once the weights are released. Anthropic CEO Dario Amodei specifically contends that this lack of control makes open-weight models harder to secure.

Conversely, advocates for open-weight models argue that this lack of centralized control is a primary strength, ensuring that no single corporation can dictate who uses the technology or how it is applied. Researchers, including former White House tech adviser Suresh Venkatasubramanian, emphasize that open-weight models are essential for scientific discovery and academic research because they can be modified and studied more freely than closed-API systems.

Model Distillation and Intellectual Property

A central point of contention is "distillation"—the process of training a new model using the outputs of another company's existing model. OpenAI and several Trump administration officials have warned against this practice, with U.S. Trade Representative Jamieson Greer characterizing Chinese distillation as a form of intellectual property (IP) theft.

If substantiated, these allegations suggest that Chinese providers used American models to develop comparable systems at a lower cost, then released subsidized versions into Western markets to undercut U.S. labs. However, critics warn that labeling distillation as IP theft could lead to de facto restrictions on all open-weight models, even those developed independently.

Regulatory Capture and Market Competition

Critics and industry observers suggest that the push for regulation is an attempt at "regulatory capture," where established leaders use safety concerns to create barriers to entry for smaller competitors.

Key points of friction include:

• Lack of Standardized Safety Metrics: There are currently no detailed safety standards for U.S. models, leaving the administration to define what "safe" looks like while incumbents lobby for the definitions.
• Divergent State-Level Approaches: While aligning federally, OpenAI and Anthropic differ on state regulation. For example, OpenAI endorsed a Massachusetts frontier safety bill that lacks the strict independent testing and civil penalty requirements supported by Anthropic.

Community Perspectives and Counterpoints

Discussion among technical communities highlights a deep skepticism regarding the motivations of frontier labs. Many argue that the push for regulation is driven by profit margins rather than safety.

"These companies have never actually cared about ensuring safe AI, regulation, or ensuring it benefits all humanity. It has always, always, been about control and their profit margins."

Other contributors point out the irony of companies trained on massive amounts of scraped public data now claiming that distillation—which is essentially querying an API—is a violation of IP. Some argue that the U.S. should instead embrace a new copyright policy that indemnifies labs while ensuring that the resulting knowledge fuels further innovation for everyone.

Furthermore, there is a concern that over-regulating open weights in the U.S. will simply drive users toward Chinese alternatives that are "good enough," potentially mirroring the trajectory of the electric vehicle market where Chinese commercial viability has outpaced Western counterparts.