|
Artificial intelligence company Anthropic has clarified that it does not support a blanket ban on open-weight AI models, with CEO Dario Amodei arguing that policy efforts should instead focus on preventing advanced AI chips from reaching authoritarian regimes, curbing industrial-scale model distillation, and mandating safety testing for powerful AI systems.
In a blog post titled "Our position on open-weights models", Amodei said speculation that Anthropic wanted to prohibit open-weight models was incorrect. "Anthropic has never advocated for a ban on open-weights models," Amodei said, adding that, "Open-weights models that don't have dangerous capabilities are a public good: they don't cost anything besides the compute needed to run them, and they provide value to businesses, developers, and researchers." The clarification comes amid an industry debate over open-weight AI models, particularly those developed in China, following reports that some US officials were considering restrictions on their use. Several major technology companies have backed an open letter supporting open-weight AI, while Anthropic's absence from the initiative had fuelled criticism that it favoured tighter restrictions. Amodei said protectionist bans would not address what he described as the two biggest national security risks: authoritarian governments developing more capable AI systems than the United States, and the misuse of advanced AI models for cyber or biological attacks. Instead, he proposed three policy measures -- tightening controls on exports of advanced AI chips and chipmaking equipment to China, cracking down on industrial-scale AI model distillation, and requiring mandatory safety testing for sufficiently capable AI models regardless of whether they are open or closed. Amodei said, "We have not and are not advocating for a ban on open-weights models as a category. We should instead focus on keeping powerful chips out of authoritarian hands, stopping industrial-scale distillation, and requiring safety testing of all sufficiently capable models, open and closed." While expressing support for many of the benefits highlighted by advocates of open-weight AI, Amodei said he disagreed with claims that open-weight models inherently improve AI safety, arguing that such assumptions should be validated through rigorous testing rather than accepted in advance. (ANI)
|
