admin The coordinated appeal was organised by the newly formed Little Tech Association and sent to President Donald Trump and senior administration officials, including Commerce Secretary Howard Lutnick and White House science adviser Michael Kratsios. The signatories include startup founders, investors and technology organisations such as Y Combinator and privacy-focused software company Proton.
Their intervention has brought a widening dispute within the US technology industry into public view. Smaller companies argue that downloadable models from developers such as Moonshot AI and Alibaba provide a low-cost alternative to proprietary systems controlled by large American laboratories. Restricting them, they say, could concentrate greater power in companies including OpenAI and Anthropic.
Open-weight models make their underlying numerical parameters available for download, allowing businesses and researchers to run, modify and fine-tune systems on their own infrastructure. This differs from closed models that are accessed through subscription services or application programming interfaces controlled by their developers.
The startups contend that local access gives them greater control over costs, privacy and product design. It can also allow companies to serve customers that are unwilling or unable to transmit sensitive information to an external provider.
A ban would force many developers to migrate to more expensive commercial services, rebuild products or abandon projects. Suhail Doshi, founder of AI search company Particle, has warned that hundreds of startups could fail immediately if access to widely used Chinese models were removed.
The campaign follows intensifying scrutiny of Moonshot AI after the release of its Kimi K3 model. The system attracted attention for combining strong performance with comparatively low operating costs and an open-weight distribution strategy.
US officials and technology executives have raised concerns that Chinese developers may have trained models using outputs generated by proprietary American systems. The process, known as distillation, allows a smaller model to learn from responses produced by a larger one. Distillation is widely used across the industry, although disputes arise when it involves alleged breaches of service agreements, unauthorised extraction or intellectual-property theft.
Anthropic has pressed Washington to take a tougher approach towards foreign companies suspected of exploiting American models. The administration has also pledged to protect US technology from adversarial states while preserving domestic innovation.
No formal plan to place Chinese AI developers on the Commerce Department’s Entity List had emerged when the startup letter was delivered. Such a designation could impose licensing requirements and restrict access to US technology, services or commercial relationships.
The Little Tech Association is calling for targeted enforcement against specific misconduct rather than country-wide or technology-wide prohibitions. Its executive director, Harry Godfrey, said policymakers should use a “scalpel rather than a sledgehammer” when addressing security and intellectual-property concerns.
The argument has gained support beyond the startup community. A separate coalition involving Microsoft, Meta, Nvidia, IBM, Dell Technologies, Palantir, Mozilla and other organisations has urged Washington to preserve open-weight development. Supporters maintain that access to model weights encourages independent testing, wider adoption and faster technological improvement.
Nvidia chief executive Jensen Huang has argued that open systems strengthen the reach of a country’s technology ecosystem. Meta has similarly promoted downloadable models as a way to prevent artificial intelligence from being controlled by a small group of companies.
Critics counter that open-weight systems can be difficult to recall or disable after release. Safety restrictions may be removed through fine-tuning, while advanced models could assist cyberattacks, disinformation campaigns or the development of dangerous biological and chemical capabilities.
Security specialists therefore favour stronger evaluations before the most capable models are published. Proposed safeguards include tiered release standards, identity checks for access to powerful versions, monitoring of computing infrastructure and penalties for developers that obtain protected data unlawfully.
The dispute also reflects a broader strategic challenge for Washington. Export controls on advanced semiconductors have limited China’s access to leading chips, but they have encouraged Chinese laboratories to improve efficiency and place greater emphasis on open technology. Models developed under those constraints are now being adopted by American researchers and companies seeking lower inference costs.
