晨间信号Morning Signal
《科技生活新闻》 2026年10月10日 · 中文解读

Anthropic向网络安全团队扩大先进AI模型访问权限

ANTHROPIC EXPANDS ADVANCED AI ACCESS FOR CYBERSECURITY TEAMS
约 6 分钟在小程序里点播,20 到 60 分钟做好
这篇讲什么
Anthropic于10月6日宣布扩大Cyber Verification Program,将此前的验证项目与Project Glasswing合并,为经过验证的网络安全专业人员提供Claude Opus 5.5、Sonnet 5.5和Mythos 5.1的三级访问权限。新计划分为防御访问、红队访问和专用访问三个层级,分别对应事件响应、授权渗透测试和航空、电网、电信及银行间转账等敏感系统的安全测试。Anthropic称在十项模拟多阶段安全挑战中,无权限时所有尝试均在首个提示处被拦截,而红队访问下模型成功完成34次尝试。
原文开头
Anthropic is widening access to its most capable AI models for verified cybersecurity professionals, giving more organizations tools to investigate threats, examine malicious software, and test their defenses with fewer automated restrictions. Announced on October 6, the expanded Cyber Verification Program brings together Anthropic’s previous verification program and Project Glasswing. It introduces three access levels covering Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1. The change addresses a practical difficulty in AI-assisted security work: investigating a vulnerability can involve techniques that also resemble an attack. Restrictions designed to prevent misuse can therefore interrupt legitimate research, including attempts to understand whether a software weakness can actually be exploited. …
摘自《科技生活新闻》(Techlife News)2026年10月10日。仅引用开头一小段供了解文章,版权归原刊所有,全文请阅读原刊。
晨间信号小程序码
微信扫码,在小程序里听完整版
不用登录先听一篇 · 或在微信搜索小程序 晨间信号
同期其他文章