Anthropic CEO Rejects Full Open Weights Despite Internal Employee Pushback

Dario Amodei defends restricted open weights citing security risks, while internal staff advocate for broader openness. Tension highlights ongoing debate over AI model accessibility versus safety protocols.

Woofun AI reports that Anthropic CEO Dario Amodei addressed the open-source controversy, stating the company does not support a complete ban on open weight models. He acknowledged that such models can reduce costs, promote competition, and facilitate enterprise deployment.

However, Amodei emphasized caution regarding strong models, noting that public weights cannot be retracted and security restrictions can be removed. He advocated for restricting advanced chip flows to China, combating industrial distillation, and mandating security testing for high-capacity models before release.

Following the statement, Shaun, an Anthropic post-training research fellow, publicly expressed disagreement. He thanked colleagues who worked with him to promote open weights. Shaun’s comments suggest opposition not to the phrase "not a complete ban," but to the continued restriction of cutting-edge model open-sourcing. Anthropic has not signed the open letter supporting open-sourcing nor committed to releasing Claude weights openly.