The Trump administration is excluding open-weight AI models from its voluntary safety testing framework, delivering a blow to closed-source advocates like Anthropic's Dario Amodei. The decision comes during a closed-door White House meeting on August 4, 2026, as Chinese rivals like Moonshot AI and DeepSeek gain rapid ground. While U.S. officials accuse Chinese firms of executing massive "distillation attacks" to copy American models, the administration is choosing to spare open-weight systems to avoid strangling domestic innovation and growth.
Open-weight AI model exemption
- ▪The White House is excluding open-weight AI models from its voluntary framework to test advanced AI capabilities
- ▪The White House delivered the decision to exempt open-weight models during a closed-door meeting on August 4, 2026, with firms including OpenAI, Anthropic, and Google
- ▪The White House AI framework defines a covered frontier model as closed-source with state-of-the-art capabilities and national security risks
Chinese AI distillation attacks
- ▪Anthropic informed the Senate Banking Committee in June 2026 that Alibaba's Qwen lab conducted the largest known distillation attack against it
- ▪Anthropic traced more than 3.4 million exchanges with its models to Moonshot AI routed through hundreds of fake accounts by February 2026
- ▪White House Science and Technology Policy Director Michael Kratsios accused Moonshot AI of improperly extracting data from Anthropic's Fable model through distillation
Distillation technique versus fraud
- ▪A distillation attack involves obtaining unauthorized access to a competitor's product through deception, such as fabricated identities and evasion infrastructure, to copy it
- ▪AI distillation is an industry-standard training method where a smaller student model is trained on the outputs of a stronger teacher model
National security risks
- ▪A Booz Allen study found that three out of four tested Chinese AI models generated software with more hidden vulnerabilities than their United States counterparts
- ▪Chinese generative AI tools are legally required by China's national cybersecurity standards committee to uphold the country's core socialist values
Economic competition displacement
- ▪Chinese open-source AI models are expanding globally and now account for the majority of global AI workloads
- ▪DeepSeek V4 Flash is priced 60% lower per task than OpenAI's GPT-5.6 Luna while scoring only one point behind it on the Artificial Analysis Intelligence Index
U.S. policy enforcement framework
- ▪Treasury Secretary Scott Bessent proposed creating an independent regulatory agency, similar to the Financial Industry Regulatory Authority, to oversee AI safety reviews
- ▪Treasury Secretary Scott Bessent threatened sanctions against Chinese AI labs found to have built their models on intellectual property theft
Story comments
Loading comments…