返回首页
ai-security
找到 3 篇关于此标签的文章
ai-securityevaluation
当 Cyber Eval 被攻破：在线评测的结构性两难
Bloomberg 8 月 25 日报道, 主要 AI labs 正在就是否把 cybersecurity benchmark 联网进行争论, 起因是近期 model hack 事件。支持者说联网能提升 accuracy, 反对者担心 integrity 被破坏。这件事表面是评测问题, 实际是整个
8月25日2 分钟
claudeanthropic
Claude Mythos Is Real: What Restricted AI Means for Builders
Anthropic's most powerful model stays locked up — here's how to work with what's available now.
4月8日2 分钟
claude-codeprompt-injection
Claude Code Security Update: What Solo Devs Need to Know
Anthropic patches Claude Code against prompt injection — here's how to keep your AI dev workflow secure.
4月7日2 分钟