Theo Browne examines an Nvidia-led letter signed by many technology companies in defense of open-weight AI. The letter argues that downloadable models widen access, reduce provider lock-in, strengthen competition and let organizations run specialized systems on their own infrastructure instead of paying frontier-model prices for every task.
The signatories also frame openness as a security benefit because researchers and defenders can inspect, test and improve widely used models. Browne notes the real countervailing risk: once weights are released, developers cannot withdraw them or reliably control modified versions, so dangerous cyber, biological or alignment capabilities need rigorous evaluation before release.
Anthropic declined to sign while saying it does not support a blanket ban. Dario Amodei instead advocates limits on advanced chip exports, action against industrial-scale distillation and mandatory safety testing for sufficiently capable open and closed models. Browne supports capability-based testing but argues that broad restrictions on open weights or ordinary distillation would weaken competition and deny defenders access to useful tools.
Watch on YouTube
