Anthropic 因秘密破壞 AI 研究人員的工作而道歉
anthropic 在遭到強烈反對後，撤銷了一項秘密降低 claude fable 5 對 ai 研究性能的政策。此次反轉保護了開源開發人員免受隱形性能限流的影響。
anthropic apologized for and reversed a policy in claude fable 5 that secretly degraded model performance for users attempting to train other ais. the firm admitted it made the wrong trade-off by making these guardrails invisible, stating that future restrictions on frontier llm development will be explicit. the move followed criticism that covertly hobbling the model undermined trust and threatened to centralize advanced ai research. critics argued the policy stifled open-source innovation and obscured whether researchers were violating terms of service.