Anthropic Just Documented Eight Months of AI Attacks.
中文摘要
Anthropic记录了八个月的AI攻击，这些攻击利用技能文件、工具服务器和子代理等通用架构，针对其Haiku、Sonnet和Opus模型进行。
English Summary
Anthropic documented eight months of AI attacks on its Haiku, Sonnet, and Opus models using accessible scaffolding like skill files, tool servers, and subagents.
Original Excerpt
Every case ran on Haiku, Sonnet, and Opus. The uplift came from scaffolding anyone can download: a skill file, a tool server, subagent… Continue reading on Medium »