Anthropic has warned that unfavorable copyright court rulings could force AI firms to destroy existing models and retrain them from scratch, posing massive financial risks.