Anthropic 红队：GLM-5.3 在 4% 试验中实现完整控制流劫持
A quote from Anthropic Frontier Red Team
Anthropic Frontier Red Team 在内部 Binary Exploitation 基准的 100 个随机任务上评测多个模型，GLM-5.3 在 4% 的试验中实现完整控制流劫持，Claude Mythos Preview 为 6%。该团队称，此前的 Claude Opus 4.6 和 GLM-5.2 在这些任务中均未成功，说明一个有意义的能力门槛已被跨过。这段引文由 Simon Willison 于 2026 年 9 月 29 日收集发布。
推荐理由
AI 评分 71
Anthropic 红队用内部 Binary Exploitation 基准对比 GLM-5.3 与 Claude Mythos Preview，给出可对照的成功率数字。
主题
正文
当前显示原文，中文全文译文尚未就绪。
29th September 2026
We evaluate several models on 100 tasks from the [internal Binary Exploitation benchmark] (selected at random), and find that GLM-5.3 develops full control flow hijacks in 4% of the trials; Claude Mythos Preview did so in 6%. Although GLM-5.3 performs below Claude Mythos Preview here, a meaningful threshold has clearly been crossed: earlier models, like Claude Opus 4.6 and GLM-5.2, do not succeed in any of them.
— Anthropic Frontier Red Team , GLM-5.3 and the spread of advanced cyber capabilities
This is a quotation collected by Simon Willison, posted on 29th September 2026 .