Anthropic红队评测：GLM-5.3在二进制漏洞利用任务中成功率达4%
Quoting Anthropic Frontier Red Team
评分 64
导读
Anthropic前沿红队从内部二进制漏洞利用基准中随机抽取100项任务评测多个模型，GLM-5.3在4%的试验中实现了完整的控制流劫持，Claude Mythos Preview为6%。虽然GLM-5.3低于Claude Mythos Preview，但此前模型如Claude Opus 4.6与GLM-5.2在这些任务中均未成功，显示相关能力门槛已被跨越。
推荐理由
Anthropic前沿红队评测显示GLM-5.3在二进制漏洞利用任务上的成功率已接近Claude预览版，读者可关注前沿模型的攻防能力变化。
主体与标签
AnthropicGLM-5.3Claude Mythos Preview#论文/研究#安全/对齐#评测/基准#Anthropic#推理
正文
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
Tags: anthropic , generative-ai , ai-security-research , glm , ai , ai-in-china , llms