Anthropic Rejects Open-Weights Ban, Keeps Safety Caveats
Anthropic says ordinary open-weight models are valuable, while arguing that the most capable systems still require targeted safeguards.
Anthropic has clarified that caution about powerful open models is not the same as supporting a blanket ban.
What happened
CEO Dario Amodei said Anthropic has never advocated banning open-weight models and described models without dangerous capabilities as useful public goods.
At the same time, he argued that the most capable systems can create national-security risks. Anthropic favours targeted controls around advanced chips, large-scale model replication and safety testing instead of broad restrictions on developers using open models.
Why it matters
Open weights let developers download, customise and run a model independently. That supports competition and local deployment, but it also removes many of the controls available when a model is accessed only through an API.
Anthropic’s position tries to separate ordinary open innovation from frontier-level risk rather than treating every downloadable model alike.
The bigger picture
The policy fight is shifting from “open versus closed” toward where capability thresholds should sit and who measures them. That distinction matters commercially: overly broad rules could protect large closed-model providers, while weak rules may fail to address genuinely powerful systems.