AINEWS 2026-10-02 中/EN Search
2026-10-02 UTC+8
INDEPENDENT PERSPECTIVES.中/EN
← 返回事件实质进展 09月30日 18:04
STORY IN FOCUS

Anthropic 评估 GLM-5.3：可自主构建漏洞利用链

Anthropic 的隔离环境测试显示，GLM-5.3 在 ExploitBench 的 410 次尝试中完成了 50 次端到端漏洞利用，与 Claude Mythos Preview 的 56 次接近。其模拟恶意请求测试中，简单方法让模型在 64% 至 100% 的情况下继续响应；这些结果不能直接等同于真实攻击表现。

2 篇报道2 个来源版本 3

报道时间线 按来源发布时间排列

量子位 AI 报道 AI 辅助摘要 同事件 机器翻译
编辑优先级 60/100
Anthropic report warns of GLM-5.3 cyber capabilities and safeguard risks

Anthropic’s report says GLM-5.3 succeeded in 50 of 410 ExploitBench attempts, compared with 56 for Claude Mythos Preview. It also says researchers got the model to continue with attacks in 92% of simulated tests by prefilling its reasoning, and calls for independent safety testing.

推荐理由：报告同时给出漏洞利用测试结果和模拟测试中的防护绕过比例，为评估模型能力与滥用风险提供了具体依据。
同一事件，精选展示《Anthropic 评估 GLM-5.3：可自主构建漏洞利用链》
Anthropic 研究成果 AI 辅助摘要 精选 机器翻译
编辑优先级 80/100
Anthropic assesses GLM-5.3’s ability to build exploits autonomously

In isolated tests, Anthropic found that GLM-5.3 completed end-to-end exploits in 50 of 410 ExploitBench attempts, compared with 56 for Claude Mythos Preview. In simulated tests involving malicious requests, simple techniques led the model to engage 64% to 100% of the time; the results do not directly establish how it would behave in real-world attacks.

推荐理由：同一评估同时给出端到端漏洞利用成功次数和防护绕过比例，并说明后者来自模拟环境。