Anthropic's chief executive said lower-risk open models are a public good and denied advocating a general ban, while preserving the company's concern about releasing frontier systems with dangerous capabilities. Anthropic CEO Dario Amodei wrote Monday that the company had never advocated a blanket ban on open-weight AI models.
Amodei described models without dangerous capabilities as a public good while maintaining that the most capable systems can create cybersecurity, biological or other misuse risks. Anthropic remained absent from a technology-industry letter signed by Nvidia, Microsoft, Meta, Google, OpenAI and other organizations warning Washington against premature restrictions on open weights.
Nvidia also launched the Open Secure AI Alliance, whose members called for open research and regulatory support for models they described as defensive assets. The dispute intensified as Chinese open-weight models improved and gained users, prompting competing claims about national security, competition, scientific access and regulatory capture.
No federal law or final administrative rule establishing a general ban or a capability threshold for open-weight models had been adopted.
Open-weight models distribute parameters that can be run or modified outside the provider's hosted service, though the term does not always include training data or fully reproducible source code. Infrastructure and platform companies can benefit when more models increase demand for chips, cloud services and developer tools, while frontier laboratories rely heavily on controlled commercial access.
Capability-based regulation requires measurements that remain reliable across model modification, fine-tuning and new tools, and the industry has not agreed on a single binding test. As of the edition cutoff, Anthropic's statement clarified its position but did not provide a complete legislative proposal or a universally accepted dangerous-capability threshold. The next scheduled factual records include federal language distinguishing open models by capability and additional signatories or revisions to the industry coalition.