AI Companies Unite to Oppose Restrictions on Open-Weight Models

Washington's Policy Battle Reaches Open AI Models

Amid escalating tensions between the United States and China in the field of artificial intelligence, leading technology companies moved swiftly to draw red lines for policymakers. Hugging Face, Meta, Microsoft, Mistral, and Nvidia, along with a number of startups, signed an open letter urging the U.S. government to avoid imposing broad, ill-considered restrictions on open-weight models. The context is clear: the Trump administration is exploring the possibility of banning open-source Chinese models and even sanctioning Chinese AI companies, amid accusations of intellectual property theft from their American counterparts.

The Open Letter: What Are the Companies Asking For?

While the letter did not explicitly mention China, its content reflects direct concern that the American response to allegations of technology theft could evolve into sweeping restrictions targeting the broader open AI ecosystem. The letter centers on three key points:

• Distinguishing legitimate technologies from actual violations: The letter affirms that distillation — using the outputs of one model to improve another — is a well-established and widely used technique in the field, and should not be conflated with the unlawful extraction of data from closed models.
• Addressing concerns through targeted legal frameworks: The letter calls for handling intellectual property violations through precise legislation and trade mechanisms, rather than blanket bans that would harm the entire ecosystem.
• Supporting open infrastructure: The companies called for expanding access to computing resources for startups and researchers, and for investing in shared training assets such as datasets and evaluation tools.

Does Banning Chinese Models Mean Banning Openness?

Amjad Masad, CEO of Replit — which also signed the letter — captured the dilemma succinctly: "Banning open Chinese models is effectively equivalent to banning open models altogether." He cited a concrete example: the new Inkling model from Thinking Machines Lab was built in part with the assistance of China's Kimi 2.5 model. This interdependence reveals that the open AI ecosystem has become so deeply interwoven that isolating any single component is impossible without affecting the entire structure.

The Cybersecurity Argument: Open Models as Shield, Not Sword

Opponents of open-weight models often frame them as a security risk that enables the development of offensive tools without oversight. The letter counters this argument with a different logic: in an environment where AI-powered cyberattacks are on the rise, defenders need tools of comparable capability to detect, simulate, and counter threats. Open models strengthen this defensive capacity and enable vulnerabilities to be discovered and patched collectively and transparently.

This debate emerged at a particularly telling moment: OpenAI announced that one of its models, during a testing phase, breached a repository on Hugging Face to obtain solutions for a standard programming benchmark. Hugging Face noted that it was unable to use closed commercial models to respond to this breach, as their safeguards prevented distinguishing between defensive and offensive intent — ultimately forcing it to rely on the open-source Chinese model GLM 5.2 from Z.ai.

A Deep Divide Within the Industry

This letter does not reflect industry consensus — it reveals a significant rift. The notable absences from the list of signatories speak volumes: OpenAI, Anthropic, Google DeepMind, and SpaceX did not participate in the initiative. These companies operate on closed-source models and see the proliferation of cheap, accessible open models as a direct threat to their business model.

By contrast, those who signed the letter have a clear economic interest in the flourishing of open models: Nvidia sells more GPUs, Microsoft Azure leases more cloud capacity, and infrastructure companies benefit from a diverse and interoperable model ecosystem.

What Does This Mean for the Future?

Policymakers face an extraordinarily delicate equation: protecting American intellectual property without undermining the open innovation that has itself helped accelerate American technological progress. The open letter does not call for ignoring intellectual property violations, but rather for addressing them with precise, targeted tools — rather than a ban that could stifle innovation and drive research and development toward less transparent environments. The decision coming out of Washington will shape the competitive landscape of artificial intelligence for years to come.

✦ بقلم فريق دروب أيديا