---
id: 20260930-T0-03
title: "Anthropic 红队测试：GLM-5.3 在4%试验中实现完整控制流劫持"
title_en: "Anthropic Red Team: GLM-5.3 Achieves Full Control Flow Hijack in 4% of Trials"
url: https://ai.daily.yangsir.net/daily/20260930-T0-03
issue_date: 2026-09-30
publish_date: 2026-09-29T22:20:28.000Z
category: research
source_name: "Simon Willison"
source_url: https://simonwillison.net/2026/Sep/29/anthropic-frontier-red-team/
---

# Anthropic 红队测试：GLM-5.3 在4%试验中实现完整控制流劫持

Anthropic Frontier Red Team 从内部 Binary Exploitation 基准中随机抽取 100 个任务评估多个模型。GLM-5.3 在 4% 的试验中实现了完整控制流劫持，Claude Mythos Preview 为 6%。GLM-5.3 在该项上低于 Claude Mythos Preview，但仍表现出有意义的攻破能力。这组数据来自 Simon Willison 的引用，说明前沿模型在二进制漏洞利用上的能力差距正在缩小。

## English Version

**Anthropic Red Team: GLM-5.3 Achieves Full Control Flow Hijack in 4% of Trials**

Anthropic's Frontier Red Team evaluated several models on 100 randomly selected tasks from its internal Binary Exploitation benchmark. GLM-5.3 achieved full control flow hijacks in 4% of trials; Claude Mythos Preview did so in 6%. While GLM-5.3 trails Claude Mythos Preview here, it still shows meaningful exploit capability. The figures, quoted by Simon Willison, suggest the capability gap between frontier models on binary exploitation is narrowing.

---

**来源**：[Simon Willison](https://simonwillison.net/2026/Sep/29/anthropic-frontier-red-team/)

**详情页**：https://ai.daily.yangsir.net/daily/20260930-T0-03

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*