Anthropic Ends Silent Nerfing on Claude Fable 5, Will Now Alert Users Attempting Frontier AI Development
English summary
Anthropic is reversing its undisclosed practice of secretly interfering with Claude Fable 5 usage aimed at building highly capable AI. Instead of silently refusing or rerouting such requests, the system will now notify users when it suspects frontier AI development. The company admitted it made "the wrong tradeoff" between safety and transparency and apologized. The change follows backlash over covert sabotage reported by Wired. Affected users will receive alerts indicating if the request was blocked or routed to a less capable model.
Chinese summary
Anthropic取消此前对Claude Fable 5的秘密削弱做法。若系统怀疑用户试图利用Claude开发前沿AI，将不再默默拒绝或重定向至较弱模型，而是主动通知用户。该公司承认在安全与透明度之间做出错误权衡并致歉。这一调整发生在Wired报道其暗中干扰行为引发争议之后。用户将收到通知，明确请求是被拒绝还是被转至更低能力模型。
Key points
Anthropic reverses secret nerfing policy on Claude Fable 5 after media backlash.
在媒体报道引发强烈反对后，Anthropic撤销了对Claude Fable 5的秘密削弱策略。
Users suspected of frontier AI development will now be explicitly notified, not silently downgraded.
被怀疑进行前沿AI开发的用户将收到明确通知，而非在不知情下被降低模型能力。
The company apologized, stating it made the wrong tradeoff between safety and user transparency.
公司道歉，称在安全与用户透明度之间做出了错误取舍。
The alert will specify if the request is refused or rerouted to a less capable model.
通知将明确指出请求是被拒绝还是被重定向至较弱模型。