“Anthropic's position against open science for its models is "anti-safety" because broader community involvement in AI development leads to safer outcomes.”