Anthropic chose not to sign an industry-backed open AI letter spearheaded by NVIDIA CEO Jensen Huang, despite sharing the broader view that open-weight AI models should remain legal. The company said it shares similar views on open-source AI but disagrees about a specific provision concerning model distillation. A technique that has become central to the ongoing debate over AI innovation, intellectual property and national security.
The clarification came after Huang's letter gained widespread industry support, with companies including OpenAI, Google, Microsoft, Meta and AMD signing the document. Anthropic and Amazon, however, remained notable absentees, prompting questions about their stance on open AI development.
Anthropic highlighted that it has never advocated for banning open-weight AI models. Rather, the firm is concerned about a certain provision in the letter that calls for prohibiting US policymakers from labeling model distillation as misappropriation.
Model distillation is a process in which a developer trains a small model using outputs generated by a larger, more sophisticated system. Although model distillation is a common practice that enhances efficiency and saves computing expenses, it has increased the possibility of companies replicating frontier AI models at lower costs.
According to Anthropic, the large-scale distillation of models, especially that claimed to be conducted by state-sponsored parties, should be viewed as a national security issue rather than ordinary AI development.
The dispute occurs against the backdrop of growing criticism of China's fast-developing AI sector. Anthropic has already claimed that some Chinese AI firms could be employing distillation methods to reduce their lag behind American rivals through advanced technology, while such technology remains unavailable to them.
Anthropic believes that industrial-scale distillation practices should be banned, while researchers should still be able to develop open-weight AI systems responsibly. Anthropic has also appealed to strengthen export control on advanced AI chips and conduct safety tests of capable AI models before releasing them publicly.
Also Read: Nadella Slams AI Limits in Anthropic Fable, Calls for Openness
Policymakers are increasingly engaged in complex policy debates about the future of artificial intelligence. It is clear that while most tech firms recognize that open-weight models are essential to drive innovations, there are conflicting views on the regulation of techniques that could allow rival firms to develop AI tools.
In his letter, Jensen Huang calls for fewer limitations on the availability of downloadable AI models, increased funding for publicly available datasets, and access to computing power by startups. On the other hand, Anthropic feels the need for measures to avoid AI distillation as a potential route for massive abuse of intellectual property and geopolitical threats.
Ongoing government efforts to outline an AI policy are expected to be characterized by debates on openness, security, and intellectual property. Anthropic's refusal to endorse the letter shows that firms that support open AI can take different stances on how evolving technologies can be controlled.