编辑精选 · @testingcatalog · 2026/10/30 02:45
Anthropic 发布 Claude 自我评估功能
Anthropic introduced self-assessment, a new technique that lets Claude give each conversation a preliminary safety assessment right after it responds. > On Clau
X 账号 @testingcatalog 报道，Anthropic 推出自我评估技术，让 Claude 在响应后即时进行初步安全评估。数据显示该机制能标记 78% 潜在有害对话及 94% 高优先级对话，且成功攻击消耗的生成令牌约为无保障时的四倍。
精选理由：X 账号报道了 Anthropic 发布 Claude 自我评估功能及安全数据
Anthropic introduced self-assessment, a new technique that lets Claude give each conversation a preliminary safety assessment right after it responds.
> On Claude Fable 5, self-assessment flags 78% of the conversations that a stronger, prompted classifier flags as potentially harmful, and 94% of the high-priority ones, at a 1% false-positive rate.
Anthropic also shed light on how self-assessment and prompted classifiers help block attacks.
> "Each successful attack costs roughly four times as many generated tokens as with no safeguards."
We already live in a world where your security depends on the compute allocated to defend the systems you rely on.