Anthropic backtracked this week after a developer backlash, saying it will stop quietly degrading Fable 5's answers to questions about frontier AI development and will instead route those requests to a weaker model, Opus 4.8, while telling developers, and it apologized. The company still limits its most powerful public model for some AI development work, citing safety and a goal of keeping foreign adversaries from eroding U.S. leadership in AI and chips, but Business Insider argues the curbs also protect Anthropic from distillation, where rivals harvest a model's outputs to train cheaper competing systems. The piece notes the limits apply broadly to anyone building rival models, Western open-source developers included, and cites data putting open models at roughly 90% of closed-model performance at a fraction of the cost, with Xiaomi's open MiMo undercutting Fable 5 by more than 20x per token.
