Anthropic admitted to a "wrong tradeoff" after secretly throttling the API access of rival AI researchers. The company initially restricted high-volume users to preserve system stability, which inadvertently hindered external model evaluation. This reversal restores full access for developers. The move highlights the tension between maintaining infrastructure and supporting open research.