Open-source AI is heading down a dangerous path—this past week, Silicon Valley has been fiercely divided over open weights. Nvidia @nvidia led an open letter arguing that Washington should not restrict open-weight models. Over the weekend, signatories have grown to 50, with OpenAI @OpenAI, Google, and SpaceX joining in succession. The only major player yet to weigh in is @AnthropicAI, whose CEO says open-source AI is heading down a dangerous path. Dario Amodei @DarioAmodei stated that when you control and deploy a model, you can continuously monitor how it’s being used; even if it’s misused at some point, you can step in and alter what it’s willing to do. But once a model is released uncontrollably into the wild, you lose that ability entirely—and everything is left to chance. He noted that open-sourcing is beneficial for small developers seeking rapid iteration, but now, when giants investing tens or even hundreds of millions of dollars to train large models release their finished products without control, it’s a fundamentally different matter.